TW201810124A - Automobile model and year identification system and method for easily and quickly obtaining automobile model information and year information through image recognition - Google Patents

Abstract

The invention discloses an automobile model and year identification system and method. The system includes an image capture device, an automobile model image database, a brand acquisition module, and an information processing device. The automobile model image database establishes automobile model image samples of at least two viewing angles, and configures feature data for the automobile model image sample of each viewing angle. Then, each feature data is configured with an automobile model data and a year data. The image capture device is configured to capture images of the automobile from at least two viewing angles. The information processing device performs image recognition to analyze two feature values of the pre-processed automobile image of at least two viewing angles, and then performs a comparison based on the automobile brand information in the corresponding automobile brand database to determine two feature data having a shape similarity of about 70% or more, so as to obtain the corresponding automobile model data and the corresponding year data from the two feature data, thereby easily and quickly obtaining the automobile model information and year information through image recognition.

Description

Car model year identification system and method

The present invention relates to a vehicle model identification system and method, and more particularly to a vehicle model image identification technology that can conveniently and quickly obtain vehicle model and year information.

According to related reports, since 1995, the number of transfers of used cars in Taiwan has exceeded the number of licenses for new cars for the first time. Until 2010, only 300,000 new cars were traded, compared with 400,000 used cars. Although the used car market will be affected by the downturn in the new car market, according to statistics from the supervision unit of the Ministry of Transport, the current output value of used cars in Taiwan can reach about NT $ 100 billion a year. It can be seen that the transactions in the used car market in Taiwan It is indeed very enthusiastic. Therefore, the major car makers are aware of the business opportunities and huge market scale of the domestic used car market, so they have entered the operating ranks of the used car market.

In addition, when buying a used car, in addition to carefully observing the internal and external conditions of the vehicle, the car manufacturer also needs to identify the brand, model, and year range of the vehicle, so that it can be based on the above identification results. Check the used car price list, and then make a reasonable and correct vehicle valuation. However, perhaps senior car dealer business personnel can carry out the above identification and evaluation on the widely circulated vehicles in the market. If the vehicle is a popular model of the unpopular car; or when the unpopular brand is a model, even the senior car dealer business staff Still unable to make a reasonable and correct vehicle valuation, thus causing inconvenience and great distress to vehicle valuation identification.

In addition, when police officers search for suspicious vehicles, their identification The device can only recognize the license plate of the vehicle. As for the more difficult models and years, the identification cannot be realized. Therefore, it is even more difficult for police officers to find suspicious vehicles in large and complex metropolises. Therefore, how to develop an image recognition technology that can identify car models and years has become a technical issue that is urgently sought to be solved and challenged by those in the relevant technical field.

In order to solve the above-mentioned shortcomings, the related technical field has already proposed a "model identification method and device" as disclosed in the new patent case of the Republic of China Invention Publication No. 201040893. The patent uses the shape of a vehicle window as a feature of identification, Homogeneous transformation is used to convert vehicle images from different perspectives to the normalized coordinate system to capture normalized window images. Although the patent can identify the target vehicle through image recognition, the patent does not have deep learning algorithms. The function is built, so the classification function of the database cannot be enhanced by accumulating the annual model characteristic data of the car model, so that the success rate of the annual model identification cannot be improved.

In view of this, there has not been a patent or a paper that has a vehicle model identification with deep learning calculations and an enhanced database classification function. Based on the urgent needs of related industries, the inventors have continuously After hard research and development, a set of inventions different from the above-mentioned conventional technologies and patents has finally been developed.

The first object of the present invention is to provide a vehicle year identification system and method. The vehicle year identification information is mainly used to conveniently identify the vehicle year information of the vehicle for transaction valuation or other purposes. Identification basis of use. The technical means adopted to achieve the first purpose of the invention include an image capture device, a car image database, a brand acquisition module and an information processing device. The car model image database establishes a car model image sample with at least two perspectives, and sets feature data in the car model image sample, and then sets a car model data and An annual information. The image capturing device is used for capturing vehicle images with at least two perspectives. The information processing device analyzes the two feature values of the at least two-viewpoint vehicle images after image pre-processing through image recognition, and then compares the discriminative similarity of about 7% with the corresponding brand car database based on the brand information. More than ten two-characteristic data, and corresponding two-characteristic data to obtain the corresponding car model data and corresponding year-type data, can not easily and quickly obtain vehicle model information and year-type information through image recognition.

A second object of the present invention is to provide a vehicle model identification system and method with an automatic valuation function, which is mainly to automatically calculate a reasonable valuation value of the vehicle after obtaining the vehicle model and the annual model information, so as to shorten the time spent reading the ancient Search time required by the car market. The technical means used to achieve the second purpose of the invention include an image capture device, a car image database, a brand acquisition module, and an information processing device. The car model image database establishes car model image samples including at least two perspectives, and sets feature data in the car model image samples, and then sets one model data and one year-type data in each feature data set. The image capturing device is used for capturing vehicle images with at least two perspectives. The information processing device analyzes the two feature values of the at least two-viewpoint vehicle images after image pre-processing through image recognition, and then compares the discriminative similarity of about 7% with the corresponding brand car database based on the brand information. More than ten two-characteristic data, and corresponding two-characteristic data to obtain the corresponding car model data and corresponding year-type data, can not easily and quickly obtain vehicle model information and year-type information through image recognition. It further includes an evaluation step that is located after the car model year identification step. Each car model data and corresponding year model data in the car model image database are set with at least one price data. When the car is completed, During the year-type identification step, the information processing device outputs the corresponding price data as price information.

A third object of the present invention is to provide a vehicle for capturing a vehicle with a predetermined viewing angle and position. Vehicle image recognition system and method for obtaining vehicle image of different perspective positions to improve the success rate of vehicle model identification. The technical means adopted to achieve the third purpose of the invention include an image capture device, a car image database, a brand acquisition module and an information processing device. The car model image database establishes car model image samples including at least two perspectives, and sets feature data in the car model image samples, and then sets one model data and one year-type data in each feature data set. The image capturing device is used for capturing vehicle images with at least two perspectives. The information processing device analyzes the two feature values of the at least two-viewpoint vehicle images after image pre-processing through image recognition, and then compares the discriminative similarity of about 7% with the corresponding brand car database based on the brand information. More than ten two-characteristic data, and corresponding two-characteristic data to obtain the corresponding car model data and corresponding year-type data, can not easily and quickly obtain vehicle model information and year-type information through image recognition. It further includes a photographing displacement control unit, which includes a moving mechanism, a transmission mechanism, and two slide rails. At least two image capturing devices can be set at different positions of the moving mechanism, and the moving mechanism can be subject to the transmission. The mechanism is driven to make a reciprocating displacement from a first position to a second position along the two slide rails. When the vehicle is parked at the position between the two slide rails and the moving mechanism is located at the first position, One of the image capturing devices captures a vehicle image from one perspective of the vehicle; when the moving mechanism is moved to the second position, the other of the image capturing device captures a vehicle from another perspective of the vehicle image.

1‧‧‧ vehicle

10, 10a, 10b, 10c‧‧‧‧Image capture device

20‧‧‧Car Model Image Database

21‧‧‧brand car series database

30‧‧‧Label Acquisition Module

40‧‧‧ Information Processing Device

40a‧‧‧web server

40b‧‧‧Computer device

50‧‧‧shooting displacement control unit

51‧‧‧ mobile agency

510‧‧‧ carrier

511‧‧‧ pulley block

52‧‧‧ Transmission mechanism

520‧‧‧Motor

521‧‧‧drive gear set

53‧‧‧Slide

54‧‧‧ sash

55‧‧‧control module

56‧‧‧Position sensing module

60‧‧‧Signal transmission module

FIG. 1 is a schematic diagram of functional block implementation of a specific architecture of the present invention.

FIG. 2 is a functional block diagram of the deep learning implementation architecture of the present invention.

FIG. 3 is a schematic diagram of the image recognition of the vehicle of the present invention after being captured and uploaded to the cloud.

FIG. 4 is a schematic diagram of implementation of image recognition of a vehicle of the present invention after being captured and uploaded to a computer.

FIG. 5 is a schematic diagram of the moving mechanism of the present invention in a first position.

FIG. 6 is a schematic diagram showing the implementation of the moving mechanism of the present invention at a middle position.

FIG. 7 is a schematic diagram of the moving mechanism of the present invention in a second position.

FIG. 8 is a schematic partial cross-sectional view of a photographing displacement control unit of the present invention at a front-view angle.

FIG. 9 is a schematic cross-sectional view of the transmission mechanism of the present invention at a plan angle.

FIG. 10 is a schematic diagram of the frame selection image selected by the present invention as a feature of a vehicle headlight group and a water tank cover.

FIG. 11 is a schematic diagram of a frame selection image selected by the present invention as a feature of a vehicle side window group.

FIG. 12 is a schematic diagram of a frame selection image selected by the present invention as a feature of a vehicle rear lamp group.

In order to allow your reviewers to further understand the overall technical features of the present invention and the technical means for achieving the purpose of the present invention, specific embodiments and drawings are described in detail below:

Please refer to Figs. 1 to 3, the first embodiment for achieving the first purpose of the present invention includes at least two image capturing devices 10, a car image database 20, a brand acquisition module 30 and a Technical characteristics such as the information processing device 40. First of all, it is based on the car image database 20, which contains a number of brand car models database 21 (such as Toyota car database, Nissan car database, Mitsubishi car database, Ford car database and Mazda car series). Database, etc .; but not limited to the above brands), each brand car department database 21 establishes a plurality of vehicle image samples containing at least two perspectives, and sets a model image sample of each model Feature data, and then set a car model data and one-year data in each feature data. The at least two image capturing devices 10 capture vehicle images of at least two perspectives of the vehicle. An image recognition software is built into the information processing device 40. When the image recognition software is executed, it includes the following image processing steps in sequence:

(a) Image pre-processing steps: At least one frame is selected in advance from the vehicle image at each perspective. A frame selection image smaller than the area of the vehicle image (coordinate values of a plurality of pixels of a specific area of the frame selection image can be defined in advance; or by manual frame selection), and then remove the image other than the frame selection image, and then frame the frame Select the image for image preprocessing; for example, perform grayscale conversion processing, Gaussian smoothing, and binarization processing on the frame-selected image in sequence. In addition, it must be noted that, in order to speed up the image processing speed, the frame selection image shown in FIG. 10 is a rectangular frame selection image covering the two headlight groups and the water tank cover in front of the vehicle; or the frame shown in FIG. 11 The selection image is a rectangular frame selection image covering the window groups on the side of the vehicle; or the frame selection image shown in FIG. 12 is a rectangular frame selection image covering the two rear light groups on the rear of the vehicle.

(b) Deep learning steps: As shown in Figure 2, a deep learning algorithm with deep learning training functions is executed to enhance the classification function of the car image database 20 by accumulating feature data; specifically, the deep learning algorithm The method can be many artificial intelligence algorithms such as artificial neural network algorithms, expert system algorithms, and random forest algorithms.

(c) Car model identification step: The two feature values of the vehicle image of at least two perspectives after image pre-processing are analyzed by image recognition, and then compared with the corresponding brand car series database based on the brand information The two-characteristic data with a discriminative similarity to the two-characteristic values of about 70% or more are obtained, and the corresponding model data and corresponding year-type data are obtained from the two-characteristic data, and then processed by the information processing device and converted into a car Information and year information. When the vehicle model data obtained by the above two characteristic data are all the same, it means that the preset similarity setting value is correct, so there is no need to adjust the setting similarity setting value; otherwise, when the above two characteristics are When the vehicle model data obtained from the data are different, it means that the preset similarity setting value is too low, so you can increase the setting value of the similarity value (such as increasing the similarity value to 100). 80% or more), until the vehicle model data obtained by the above two characteristics data is the same.

In an application embodiment based on the above embodiments, the information processing device 40 may be a network server, a computer, a smart phone, or a tablet computer. As shown in FIG. 3, it indicates that the information processing device 40 is a network server 40a. The information of the line signal transmission module 60 (such as a combination of a wired or wireless communication network and a network transmission interface) is uploaded to a network server, and the server performs the above-mentioned image processing steps. In addition, FIG. 4 The information processing device 40 shown is a computer. After obtaining the vehicle image, it is transmitted to the computer 40b through a wired or wireless signal transmission module 60 (such as RS232, USB and other transmission interfaces), and the computer 40b performs the above-mentioned Image processing steps.

This embodiment is a second embodiment for achieving the second purpose of the present invention. This embodiment is substantially the same as the technical structure of the first embodiment described above. The technical difference between the two is that this embodiment further includes The evaluation step after the identification step is performed. At the time of performing the evaluation step, at least one price data is set for each car model data and corresponding year-type data in the car model image database 20. When the car model year-end identification step is completed At this time, the information processing device 40 outputs the matched price data as price information.

Please refer to FIG. 1 to FIG. 4 for a third embodiment for achieving the third purpose of the present invention. This embodiment is substantially the same as the first embodiment in the technical architecture. The technical difference between the two is that this embodiment is It further includes a photographing displacement control unit 50. The photographing displacement control unit 50 includes a moving mechanism 51, a transmission mechanism 52, and two slide rails 53. The moving mechanism 51 can be provided with at least two sets of image capturing devices 10 at different positions. The moving mechanism 51 can be driven by the transmission mechanism 52 to make a reciprocating displacement between a first position and a second position along the two slide rails 53; when the vehicle 1 is parked at a position between the two slide rails 53 and the moving mechanism 51 When it is located at the first position (as shown in FIG. 5), an image capturing device 10a captures a vehicle perspective of the vehicle 1, as shown in FIG. 5 (a forward-looking angle inclined from top to bottom), This vehicle image is shown in FIG. 10; when the moving mechanism 51 is displaced to the second position, as shown in FIG. 7, the second image capturing device 10b captures the other two perspectives of the vehicle 1 (as shown in FIG. 7 from above). A downwardly tilted rear-view angle) vehicle image. This vehicle image is shown in FIG. 12.

Specifically, please cooperate with the moving mechanism 51 shown in Figs. A carrier frame 510 which is slightly ㄇ -shaped, and two pulley sets 511 pivoted at the two ends of the carrier 510, and the two pulley sets 511 can be rotated along the two slide rails 53; the transmission mechanism 52 shown in Figs. 8 and 9 The system includes two sets of motors 520 and two sets of transmission gear sets 521, which are used to drive the two pulley sets 511 to move forward, backward, and stop along the two slide rails 53; in addition, the two shown in Figures 5-7 A frame 54 is provided between the slide rails 53 for the vehicle 1 to park and take pictures. In addition, the photographing displacement control unit 50 shown in FIG. 1 further includes a control module 55 and a position sensing module 56; this position The sensing module 56 is used to sense whether the vehicle 1 is parked in the frame 54. If the sensing result is yes, a sensing signal is output to the control module 55. After being interpreted by the control module, an image capturing device is activated. 10a, to capture a vehicle image from one perspective of the vehicle. After a preset time (about 1 to 30 seconds), the control module 55 triggers the transmission mechanism 52 to cause the transmission mechanism 52 to drive the carrier 510 from the first position. Move to the second position, and activate the second image capturing device 10b to capture the vehicle image of the vehicle 2 from two perspectives, After a preset time (about 1 to 30 seconds) has passed, the control module triggers the transmission mechanism 52 again to drive the carrier 510 from the second position to the original first position. Specifically, one of the image capturing devices 10a shown in FIGS. 3 to 5 is located on the front side of the top of the carrier 510, and the axis and plane of the lens of the image capturing device 10a (ie, the X and Y axes) have 30 An angle of ~ 70 degrees; the second image capturing device 10b is located on the rear side of the top of the carrier 510, and the lens axis and plane (ie, X and Y axes) of the second image capturing device 10b have an angle of 30 to 70 degrees . In addition, the image capturing device 10 further includes a three image capturing device 10c located on one side of the carrier. During the movement of the carrier 510 from the first position to the second position, when the carrier 510 enters the first At the middle position between the position and the second position, the control module 55 stops the transmission mechanism 52 for a preset time (about 10-50 seconds) and activates its three image capturing device 10c to capture the vehicle's A vehicle image (that is, a vehicle image shown in FIG. 11) of three viewing angles (that is, a side view angle shown in FIG. 6).

In addition, in the above-mentioned image processing steps, the present invention uses a template feature comparison method to sequentially calculate the feature values sufficient to describe the characteristics of the vehicle from the vehicle images at various perspectives. The above feature data is a feature table. Each feature table writes a specific brand and the car models and year types arranged in order. In the image processing step, the present invention uses a random forest algorithm to separate each vehicle image. Each feature value is compared with each feature table one by one. Here, the corresponding car model and year type can be obtained (such as the Toyota brand: the model is ALTI: the year is 2010-2015, etc.).

Specifically, in the image processing step of the present invention, a concept similar to the template feature comparison method is used. In order to find out the car models and year-type categories that need to be identified, and through the writing of recognition software, it is necessary to calculate enough to describe the vehicle. The value of the object characteristics of the image. This is the component characteristic value (the above-mentioned characteristic value). Then, collect a lot of model objects that are slightly different from each other (that is, the image samples of the car models), and use the random forest algorithm to perform the characteristics. Accumulate and calculate training to calculate the different models and year types that should belong to them. This method is the same as the principle of cluster analysis. Assume that there is a component set C of the image of vehicle 1. The system The characteristic values of all components must be calculated for comparison and identification. The calculation formula is as follows:

In formula (1), w and h respectively represent the width and height of the feature capture area of the vehicle 1 image, x and y are randomly generated coordinate positions, and the range is limited to the frame 1 selected image of the vehicle. P represents ( x , y ) The pixel value in the coordinates; in formula (2), all p belong to the pixel value set P , the subscript i is 0 to n , which represents the number of pixel values in the set, f is the feature binary code, and the subscript j is 1 to n ; this formula means that if the current pixel value is not equal to the next pixel value, the feature binary code is 1, otherwise it is 0; formula (3) is a formula combining feature binary codes, and the random coordinate points are according to They are arranged in order from top to bottom, separated by auxiliary dotted lines. The pixel value of the coordinate point at the top layer is p ₀ , and the bottom layer is p _6. Because p ₀ is not equal to p ₁ , f ₁ is 1, and because p ₁ and p _{2 is} equal, so f ₂ is 0, and so on. Finally, the characteristic value of the component is 100111, and F is equal to 39 through formula (3). This example shows how to calculate the feature binary code, which is described by numerical values. The feature of the element, on the other hand, if the more coordinate points are randomly generated, the more bits of the feature binary code will be.

In this embodiment, the random forest algorithm is used as the above-mentioned deep learning algorithm. The vehicle model year identification step is to vote by referring to the feature table. The person who receives the most votes is regarded as the vehicle model year category to which the component belongs. The positions are extracted from three randomly generated feature values and substituted into the feature table calculated by the car image sample. Each element corresponds to a different feature table with a different index value. After the system calculates the feature value of the vehicle 1 image According to the way of checking the table, the corresponding car models and year-type categories in Table 1 are voted, and the category with the most votes is the identification result of the component. For example, input a component and calculate its characteristics. After calculating the feature location from the first group, the characteristic value is 42, the second group is 80, and the third group is 22. The corresponding VIOS, 2010-2015 model Two votes were obtained for the year-and-year category, and one vote for the CAMRY, 2005-2010 year and year-year category. The system judged that the brand of this vehicle 1 was Toyota, the model was VIOS, and the year was 2010-2015.

The image recognition method used in the present invention belongs to the deformation of the Random Forest Algorithm (RF). In describing the characteristics of the image, multiple sets of random coordinates are generated in the image, and the images belonging to the target image are obtained by comparison and combination. Feature binary code, and the follow-up algorithm to achieve the purpose of image tracking. The method of the present invention adopts the application method of this method in part, and designs a classifier suitable for vehicle type and year type identification. In the system processing process of vehicle type and year type identification, the actual image can be used as the detection target. The measurement target establishes its characteristics, obtains the pixel value of the point from the randomly generated coordinates, sets the bit value by comparing the pixel values with each other, and uses all the obtained binary codes as the component feature values and component features of the image. After the calculation is completed, multiple sets of feature tables created from the vehicle image samples in the image database of Vehicle 1 are used to represent the component's integrated classifier (Ensemble Classifier), and finally component identification is performed.

When the forest is initially created, the invention does not use a random sampling method to establish a decision tree. Instead, it inputs all component templates (that is, car image samples) and extracts positions based on different feature values to establish different decision trees to form "features "Table forest", and finally strengthen each weak classification through training, so that the forest has a good component identification ability. Furthermore, a decision tree is based on a randomly generated set of coordinate points, which is established after inputting all the templates. A decision tree model can be defined as shown in the following formula (4):

Due to the strain identification algorithm, it is necessary to rely on the template features (that is, the vehicle image samples) to make it robust. As the feature table is required to build the template data (that is, the feature data), the larger the number, the better. The classifier is more robust. So far, each car model and year category can each have about 100 comparison car image samples, for a total of about 700 comparison car image samples. Furthermore, the concept of the random forest algorithm used in the present invention, and according to a specially designed feature value extraction method, the values representing the features are converted into a decision tree, that is, a feature table, as an integrated model and year The classifier of the type category knows from the implementation results that each feature table has a recognition ability of more than 70%, and can independently distinguish the annual model of the car model with more than 70% of the test data.

The above description is only a feasible embodiment of the present invention, and is not intended to limit the patent scope of the present invention. Any equivalent implementation of other changes based on the content, characteristics and spirit of the following claims should be It is included in the patent scope of the present invention. The structural features specifically defined in the present invention are not found in similar items, and are practical and progressive. They have met the requirements for invention patents. They have filed applications in accordance with the law. I would like to request the Bureau to verify the patents in accordance with the law in order to maintain this document. Applicants' legitimate rights and interests.

1‧‧‧ vehicle

10, 10a, 10b, 10c‧‧‧‧Image capture device

20‧‧‧Car Model Image Database

21‧‧‧brand car series database

30‧‧‧Label Acquisition Module

40‧‧‧ Information Processing Device

50‧‧‧shooting displacement control unit

52‧‧‧ Transmission mechanism

55‧‧‧control module

56‧‧‧Position sensing module

60‧‧‧Signal transmission module

Claims (10)

An automobile type identification system includes: at least two image capturing devices for capturing vehicle images of at least two perspectives of a vehicle; and a vehicle image database, which establishes a factory containing a plurality of vehicle brands Brand car series database, each brand car series database creates a plurality of car model image samples containing at least two perspectives, and sets a feature data on the model image samples of each perspective, and then in each The characteristic data is set with a car model data and one-year data; a brand acquisition module that is used to obtain the brand information of the vehicle; and an information processing device that has built-in image recognition software to execute the image recognition software The sequence includes the following image processing steps in sequence: (a) image processing steps: pre-processing image of the vehicle images of at least two perspectives in sequence; and (b) year-of-year identification step of the vehicle model: image identification analysis The two eigenvalues of the vehicle image in at least two perspectives after image pre-processing are generated, and the discriminative phase with the two eigenvalues is compared with the corresponding brand car database based on the brand information. The two characteristic data of about 70% or more are obtained from the two characteristic data, and the corresponding model data and corresponding year-type data are obtained, and then processed by the information processing device and converted into the vehicle type information and the year-type. Information. The annual model identification system as described in claim 1, further comprising an evaluation step subsequent to the annual model identification step of the model, which is each of the model information in the model image database and the corresponding The annual data is set with at least one kind of price data. When the annual model identification step of the car model is completed, the information processing device outputs the corresponding price data as price information. The car model identification system according to claim 1, further comprising a deep learning step, which executes a deep learning algorithm with deep learning training function to accumulate the feature value data to strengthen the car image database Classification function, the information processing device is selected from a network server, a computer, a smart phone, and a tablet, and the brand acquisition module is An input interface displayed on a display screen of one of the information processing devices, and the input interface can be used to input a command of a vehicle brand, and the brand information is generated after being interpreted by the information processing device. The vehicle type identification system according to claim 1, further comprising a photographing displacement control unit including a moving mechanism, a transmission mechanism, and two slide rails. Different positions of the moving mechanism can be set at least Two image capturing devices, the moving mechanism can be driven by the transmission mechanism to make a reciprocating displacement from a first position to a second position along the two slide rails when the vehicle is parked on the two slide rails When the moving mechanism is in the first position, one of the image capturing devices captures a vehicle image of the vehicle from one perspective; when the moving mechanism is moved to the second position, the second image The capture device captures a vehicle image of the vehicle from two perspectives. The year-of-year identification system for a vehicle as described in claim 4, wherein the moving mechanism includes a slightly cymbal-shaped carrier and two pulley sets pivoted at the two ends of the carrier, respectively. As the two slide rails rotate, a frame is arranged between the two slide rails for the vehicle to park and take pictures; the shooting displacement control unit further includes a control module and a position sensing module; the position sensing module The group is used to sense whether the vehicle is parked in the frame. If the sensing result is yes, a sensing signal is output to the control module. After the control module interprets it, it starts one of the image capture devices to Capture a vehicle image of the vehicle from a different perspective, and after a preset time, the control module triggers the transmission mechanism, so that the transmission mechanism drives the carrier to move from the first position to the second position, And activate the second image capturing device to capture the vehicle image from the second perspective of the vehicle. After a preset time, the control module triggers the transmission mechanism again to drive the carrier from the first The two positions return to the original first position. The vehicle identification system according to claim 5, wherein an image capturing device is located on the front side of the top of the carrier, and an axis and a plane of a lens of the image capturing device have 30 An angle of ~ 70 degrees; the second is that the image capturing device is located on the rear side of the top of the carrier, and the second is that the axis of the lens of the image capturing device and the plane have an angle of 30 to 70 degrees. The vehicle identification system according to claim 5, wherein the image capturing device further includes a three image capturing device located on one side of the carrier, and the carrier moves from the first position to the vehicle. During the movement of the second position, when the carrier enters the middle position between the first position and the second position, the control module stops the transmission mechanism for a preset time and starts three The image capturing device captures the vehicle images from three perspectives of the vehicle. An annual identification method for a vehicle model includes: providing at least two image capturing devices, a photographing displacement control unit, a car image database, a brand acquisition module, and an information processing device; wherein the photographing displacement control The unit includes a moving mechanism, a transmission mechanism, and two slide rails. The at least two image capturing devices can be set at different positions of the moving mechanism. In the vehicle image database, a brand car system including a plurality of vehicle brands is established. Database, each vehicle model database of the brand establishes a plurality of vehicle image samples containing at least two perspectives, and sets a characteristic data for the vehicle image samples for each perspective, and then sets for each of the characteristic data There is a car model data and one-year data; the transmission mechanism is used to drive the moving mechanism to reciprocate from a first position to a second position along the two slide rails when the vehicle is parked on the two slide rails When the moving mechanism is in the first position, the image capturing device is made to capture a vehicle image of the vehicle from a different perspective; when the moving mechanism is moved to the first position When it is in the position, the second image capturing device is used to capture the vehicle image of the vehicle from another perspective; the brand acquisition module is used to obtain the brand information of the vehicle; and an image recognition software is built in the information processing device. , When executing the image recognition software, the following image processing steps are sequentially included: (a) image processing steps: image pre-processing of the vehicle images of at least two perspectives in sequence; and (b) year model identification Step: Use image recognition to analyze at least the image after pre-processing The two feature values of the vehicle image from two perspectives are then compared to the corresponding brand car database based on the brand message to compare the two feature values with a discriminative similarity of more than 70% with the two feature values. Characteristic data, and obtain the corresponding model information and corresponding year-type data from the two characteristic data, and then process and convert the information processing device into vehicle model information and year-type information. The method for identifying a vehicle model according to claim 8, wherein when the image processing step is performed, at least one frame selection image smaller than the area of the vehicle image is selected in advance from the vehicle image, and then the frame selection is removed. Frame select images other than the images, and then perform frame pre-processing on the frame selected images. According to the method of identifying the model of a car model as described in Request 8, it further includes an evaluation step after the step of identifying the model of the car model, which is each of the model information in the model image database and the corresponding year At least one kind of price data is set in the formula data. When the step of identifying the model of the car is completed, the information processing device outputs the price data that matches the price information.