Vehicle automatic detection system and method based on multi-feature deep learning
By using multi-feature deep learning technology in the vehicle re-identification system, multiple key features of the vehicle are extracted and fused, the problem of poor vehicle re-identification results caused by global features in the prior art is solved, and higher recognition accuracy and robustness are achieved.
Patent Information
- Application Number
- CN202411457717.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-18
- Publication Date
- 2025-02-25
AI Technical Summary
The use of global features in vehicle re-identification results in poor results, especially when there are domain deviations between different data sets, the vehicle re-identification performance will be greatly reduced.
The automatic vehicle detection system based on multi-feature deep learning is adopted to extract key features of the vehicle (including vehicle profile features, vehicle color features, front features and rear features) and establish a neural network model for joint training, combining data enhancement technology and multi-feature fusion to improve the accuracy of vehicle re-identification.
Through multi-feature fusion and deep learning technology, the accuracy and robustness of vehicle re-identification are improved, inter-domain differences are reduced, and the adaptive ability of cross-domain vehicle re-identification is improved.
Smart Images
Figure CN119516480A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent traffic management, and in particular to a vehicle automatic detection system and method based on multi-feature deep learning. Background Art
[0002] Vehicle re-identification aims to match the same target vehicle from different camera perspectives. It plays an important role in transportation, public security and video surveillance. The vehicle re-identification task mainly includes two steps: feature extraction and similarity measurement. The traditional vehicle re-identification method mainly extracts some low-level image features manually, which has low re-identification accuracy. In recent years, with the rapid development of convolutional neural networks and deep learning, more and more deep learning models have been applied to vehicle re-identification problems. This has also led to a continuous improvement in vehicle re-identification performance.
[0003] With the emergence of big data sets, there are large domain deviations between different data sets, with different data distributions, and image styles, lighting, occlusion and other information vary greatly. When a model trained on one data set is directly applied to another data set, the vehicle re-identification performance will drop significantly. At present, vehicle re-identification based on deep learning is very mature in single feature learning, but there is relatively little research on multi-feature hybrid learning methods. Summary of the invention
[0004] The purpose of the present invention is to solve the technical problem of poor effect caused by using global features for vehicle re-identification, and to propose a vehicle automatic detection system and method based on multi-feature deep learning.
[0005] In order to achieve the above object, the present invention adopts the following technical solution: a vehicle automatic detection method based on multi-feature deep learning, comprising the following steps:
[0006] S1: Establish vehicle image dataset;
[0007] S2: extracting key features of the vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features;
[0008] S3: establishing a re-identification model, wherein the re-identification model is obtained by joint training using the vehicle contour feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature;
[0009] S4: Correcting the re-identification model, using the test data to test and correct the re-identification model in step S3 to obtain a corrected re-identification model;
[0010] S5: Acquire a real-time vehicle image, and use the modified re-identification model of step S4 to perform recognition, thereby completing vehicle re-identification.
[0011] As a further description of the above technical solution:
[0012] The vehicle image dataset establishment in step S1 includes the following steps:
[0013] S11: collecting original image data of the vehicle;
[0014] S12: Create an initial vehicle training data set;
[0015] S13: performing data enhancement processing on the initial training data set to obtain a final training data set;
[0016] S14: Using the final training data set to establish a vehicle image data set.
[0017] As a further description of the above technical solution:
[0018] The data enhancement process comprises the following steps:
[0019] S131: performing binarization processing and connected domain analysis on the training data image to obtain a first image;
[0020] S132: Perform perspective and scaling processing on the first image to obtain a second image;
[0021] S133: Obtain a pasting position according to the geometric size of the second image and the vanishing point position of the training data image, and finally paste the second image onto the training data image according to the pasting position.
[0022] As a further description of the above technical solution:
[0023] The re-identification model in step S3 is a neural network model.
[0024] As a further description of the above technical solution:
[0025] The steps of establishing the re-identification model in step S3 are:
[0026] S31: Acquire the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature in step S2;
[0027] S32: Fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature to obtain three fused features of different scales.
[0028] S33: The re-identification model is trained using fusion features.
[0029] A vehicle automatic detection system based on multi-feature deep learning, comprising:
[0030] An image acquisition module, used to acquire vehicle image data sets;
[0031] A feature extraction module is used to extract key features of a vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features;
[0032] The re-identification module is used to receive the information from the feature extraction module and fuse the received information with the real-time vehicle image for similarity matching to obtain the vehicle re-identification result.
[0033] As a further description of the above technical solution:
[0034] The re-identification module includes a fusion submodule and a correction submodule;
[0035] A fusion submodule, used for fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature to obtain fusion features of three different scales;
[0036] The correction submodule is used to correct the re-identification module.
[0037] In summary, due to the adoption of the above technical scheme, the beneficial effects of the present invention are: by collecting the initial image data of the vehicle to establish a vehicle image data set, designing a vehicle re-identification model based on a neural network model, using an easy-to-implement algorithm to extract various feature maps of the image, and then performing multi-feature fusion, and further extracting convolution features from the obtained feature maps through deep learning, so that the obtained features are more classification discriminative and the discrimination effect is improved; these feature maps are input into the re-identification model for similarity matching with real-time vehicle images to obtain vehicle re-identification results, which can improve the classification accuracy of the vehicle and improve the detection recognition rate. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 The present invention is a flowchart of a method for transmitting information in a communication network in an offline state. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] See also Figure 1 The present invention provides a technical solution: a vehicle automatic detection system and method based on multi-feature deep learning, comprising the following steps:
[0042] S1: Establish vehicle image dataset;
[0043] S2: extracting key features of the vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features;
[0044] S3: establishing a re-identification model, wherein the re-identification model is obtained by joint training using the vehicle contour feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature;
[0045] S4: Correcting the re-identification model, using the test data to test and correct the re-identification model in step S3 to obtain a corrected re-identification model;
[0046] S5: Acquire a real-time vehicle image, and use the modified re-identification model of step S4 to perform recognition, thereby completing vehicle re-identification.
[0047] The vehicle image dataset establishment in step S1 includes the following steps:
[0048] S11: collecting original image data of the vehicle;
[0049] S12: Create an initial vehicle training data set;
[0050] S13: performing data enhancement processing on the initial training data set to obtain a final training data set;
[0051] S14: Using the final training data set to establish a vehicle image data set.
[0052] The original image data is the image of the vehicle recorded at the vehicle management office, which can improve the accuracy of the original data, thereby facilitating the identification of subsequent vehicles and improving the recognition rate.
[0053] The data enhancement process comprises the following steps:
[0054] S131: performing binarization processing and connected domain analysis on the training data image to obtain a first image;
[0055] S132: Perform perspective and scaling processing on the first image to obtain a second image;
[0056] S133: Obtain a pasting position according to the geometric size of the second image and the vanishing point position of the training data image, and finally paste the second image onto the training data image according to the pasting position.
[0057] The first image is obtained by performing binarization processing and connected domain analysis on the training data image, and then the first image is subjected to perspective and scaling processing to obtain the second image. Then, the pasting position is obtained according to the geometric size of the second image and the vanishing point position of the training data image, and finally the second image is pasted to the training data image according to the pasting position, thereby improving the image quality and further improving the image recognition effect.
[0058] The re-identification model in step S3 is a neural network model. The neural network model is specifically a Y0L0V5 backbone network, and is pre-trained on the public ImageNet dataset to obtain a neural network model.
[0059] The steps of establishing the re-identification model in step S3 are:
[0060] S31: Acquire the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature in step S2;
[0061] S32: Fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature to obtain three fused features of different scales.
[0062] S33: The re-identification model is trained using fusion features.
[0063] The use of inter-domain joint training and multi-feature hybrid learning can reduce inter-domain differences, making the system more stable and more robust. The images input from different camera perspectives are trained jointly with global and local features, and after hybrid learning, image comparison and matching are performed, which improves the adaptive ability of the cross-domain vehicle re-identification model, performs vehicle re-identification on cross-domain data sets, and improves cross-domain vehicle re-identification.
[0064] A vehicle automatic detection system based on multi-feature deep learning, comprising:
[0065] An image acquisition module, used to acquire vehicle image data sets;
[0066] A feature extraction module is used to extract key features of a vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features;
[0067] The re-identification module is used to receive the information from the feature extraction module and fuse the received information with the real-time vehicle image for similarity matching to obtain the vehicle re-identification result.
[0068] The re-identification module includes a fusion submodule and a correction submodule;
[0069] A fusion submodule, used for fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature to obtain fusion features of three different scales;
[0070] The correction submodule is used to correct the re-identification module.
[0071] Working principle: By collecting the initial image data of the vehicle to establish a vehicle image dataset, a vehicle re-identification model is designed based on a neural network model, and an easy-to-implement algorithm is used to extract various feature maps of the image. Multi-feature fusion is performed, and convolution features are further extracted from the obtained feature maps through deep learning, so that the obtained features are more discriminative and the discrimination effect is improved; these feature maps are input into the re-identification model for similarity matching with real-time vehicle images to obtain vehicle re-identification results, which can improve the classification accuracy of vehicles and improve the detection recognition rate.
[0072] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A vehicle automatic detection method based on multi-feature deep learning, characterized in that: The following steps are involved: S1: Establish vehicle image dataset; S2: extracting key features of the vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features; S3: establishing a re-identification model, wherein the re-identification model is obtained by joint training using the vehicle contour feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature; S4: Correcting the re-identification model, using the test data to test and correct the re-identification model in step S3 to obtain a corrected re-identification model; S5: Acquire a real-time vehicle image, and use the modified re-identification model of step S4 to perform recognition, thereby completing vehicle re-identification.
2. The vehicle automatic detection method based on multi-feature deep learning according to claim 1 is characterized in that: The vehicle image dataset establishment in step S1 includes the following steps: S11: collecting original image data of the vehicle; S12: Create an initial vehicle training data set; S13: performing data enhancement processing on the initial training data set to obtain a final training data set; S14: Using the final training data set to establish a vehicle image data set.
3. The vehicle automatic detection method based on multi-feature deep learning according to claim 2 is characterized in that: The data enhancement process comprises the following steps: S131: performing binarization processing and connected domain analysis on the training data image to obtain a first image; S132: Perform perspective and scaling processing on the first image to obtain a second image; S133: Obtain a pasting position according to the geometric size of the second image and the vanishing point position of the training data image, and finally paste the second image onto the training data image according to the pasting position.
4. The vehicle automatic detection method based on multi-feature deep learning according to claim 1 is characterized in that: The re-identification model in step S3 is a neural network model.
5. The vehicle automatic detection method based on multi-feature deep learning according to claim 1 is characterized in that: The steps of establishing the re-identification model in step S3 are: S31: Acquire the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature in step S2; S32: Fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature, and the vehicle rear feature to obtain three fused features of different scales. S33: The re-identification model is trained using fusion features.
6. A vehicle automatic detection system based on multi-feature deep learning, characterized in that: include: An image acquisition module, used to acquire vehicle image data sets; A feature extraction module is used to extract key features of a vehicle, wherein the key features include vehicle outline features, vehicle color features, vehicle front features, and vehicle rear features; The re-identification module is used to receive the information from the feature extraction module and fuse the received information with the real-time vehicle image for similarity matching to obtain the vehicle re-identification result.
7. The vehicle automatic detection system based on multi-feature deep learning according to claim 6 is characterized in that: The re-identification module includes a fusion submodule and a correction submodule; A fusion submodule, used for fusing the vehicle outline feature, the vehicle color feature, the vehicle front feature and the vehicle rear feature to obtain fusion features of three different scales; The correction submodule is used to correct the re-identification module.