A vehicle detection method and system based on roadside parking scene

Abstract

The present invention discloses a vehicle detection method and system based on a roadside parking scene, and relates to the field of intelligent parking management. The method comprises: using a target detection framework based on a region candidate frame, first inputting an input image into a backbone network for feature extraction to obtain a feature map, and then using a region generation network to obtain a vehicle candidate frame, on the one hand, sending the vehicle candidate frame into a candidate region alignment network to perform size alignment on the input feature; on the other hand, sending the vehicle candidate frame into a pre-trained vehicle large model to extract more refined vehicle features; finally, splicing the features of the two aspects in the channel dimension, and sending them into a vehicle detection and classification network to obtain a final output, which can achieve a more accurate vehicle detection result, and can provide a more accurate judgment basis when performing a parking management task by using the roadside parking scene.

Description

Technical Field

[0001] The present invention relates to the field of intelligent parking management, and in particular to a vehicle detection method and system based on a roadside parking scenario. Background Art

[0002] In recent years, with the development of social economy, people's demand for convenience in daily travel has continued to increase, and the number of cars has continued to increase. However, the increase in the number of parking spaces is far less than the increase in the number of vehicles, resulting in difficulty in parking and disorderly parking, which has become one of the main problems that need to be solved in the development of urban transportation in my country. At present, traffic control departments in various places are also actively carrying out and promoting solutions to the problems of difficulty and disorder in parking for the majority of car owners. Especially in recent years, with the development of intelligent technology, in order to effectively alleviate this problem, traffic control departments in various places have tried and built various intelligent parking management technologies. Among them, by setting up parking spaces on the roadside, using intelligent sensing equipment to collect data such as video images, and processing and analyzing video image data through computer vision algorithms, intelligent parking management is currently a mainstream solution. At present, it is mainly achieved by installing high-position video cameras on the roadside for data collection, and then using visual algorithms for vehicle detection to achieve parking management of vehicle targets in roadside parking scenes.

[0003] Vehicle detection tasks are crucial for fine-grained vehicle analysis in roadside parking scenarios. Currently, for roadside parking scenarios, it is a scenario where parking fees are charged to consumers, so higher accuracy is required for vehicle detection. At present, some research methods extract and utilize multimodal data to solve these challenging vehicle detection scenarios, such as adding lidar, thermal infrared imaging cameras, etc. However, on the one hand, the cost of installing these devices is high, and it is difficult to achieve large-scale installation in roadside areas; on the other hand, the application of multimodal models may require higher computing power for running equipment, and currently this method is difficult to deploy on a large scale. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a vehicle detection method and system based on roadside parking scenarios, which can solve the problems of high difficulty and high cost in the existing vehicle detection based on roadside parking scenarios.

[0005] To achieve the above object, the present invention provides a vehicle detection method based on a roadside parking scenario, the method comprising:

[0006] Extracting features of the image frame by using a feature extraction model preset in the vehicle detection network, and performing feature aggregation on the feature information;

[0007] Extracting candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model in a preset vehicle detection network, and aligning the feature size of the candidate frame position information by using a feature size alignment model in a preset vehicle detection network;

[0008] Extracting refined vehicle feature information from the candidate frame position information by using a pre-trained large model in a preset vehicle detection network, and performing feature splicing on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature splicing model in the preset vehicle detection network;

[0009] By presetting the vehicle classification and regression model in the vehicle detection network, the vehicle target detection result is output according to the feature map after the feature splicing.

[0010] Furthermore, the step of extracting features of the image frame by using a feature extraction model preset in the vehicle detection network and performing feature aggregation on the feature information includes:

[0011] According to the resolution size of the original video corresponding to the image frame, the feature extraction model in the preset vehicle detection network is used to extract the features of the image frame, and the feature information between different layers of the preset vehicle detection network is fused.

[0012] Furthermore, the step of extracting candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in the vehicle detection network, and aligning the feature size of the candidate frame position information by using a feature size alignment model preset in the vehicle detection network includes:

[0013] The feature information after the feature aggregation is transformed into a feature dimension by using a candidate region extraction model preset in a vehicle detection network to obtain a classification result and candidate frame position information;

[0014] Performing bilinear interpolation processing on the candidate frame position information through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers;

[0015] Feature size alignment is performed according to image values ​​at pixel points whose coordinates are floating point numbers.

[0016] Furthermore, the step of performing feature stitching on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature stitching model preset in the vehicle detection network includes:

[0017] The feature map after the feature size alignment and the refined vehicle feature information are spliced ​​in the channel dimension through a large model feature splicing model preset in the vehicle detection network.

[0018] Furthermore, before the step of extracting features of the image frame by using a feature extraction model preset in the vehicle detection network and performing feature aggregation on the feature information, the method further includes:

[0019] Through the target detection loss function L = α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , the target detection loss function L, α1 and α2 are weight coefficients respectively.

[0020] Furthermore, the present invention provides a vehicle detection system based on a roadside parking scenario, the system comprising:

[0021] An extraction module, used to extract features of the image frame by using a feature extraction model preset in the vehicle detection network, and perform feature aggregation on the feature information;

[0022] An alignment module, used to extract candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in a vehicle detection network, and perform feature size alignment on the candidate frame position information by using a feature size alignment model preset in the vehicle detection network;

[0023] A splicing module, used to extract refined vehicle feature information from the candidate frame position information by using a pre-trained large model in a preset vehicle detection network, and to perform feature splicing on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature splicing model in the preset vehicle detection network;

[0024] The target detection module is used to output the vehicle target detection result according to the feature map after the feature splicing by presetting the vehicle classification and regression model in the vehicle detection network.

[0025] Furthermore, the extraction module is specifically used to extract features of the image frame according to the resolution size of the original video corresponding to the image frame through a feature extraction model in a preset vehicle detection network, and to fuse feature information between different layers of the preset vehicle detection network.

[0026] Furthermore, the alignment module is specifically used to perform feature dimension transformation on the feature information after feature aggregation through a candidate area extraction model preset in a vehicle detection network to obtain classification results and candidate box position information; perform bilinear interpolation processing on the candidate box position information through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers; and perform feature size alignment according to the image values ​​at pixel points whose coordinates are floating point numbers.

[0027] Furthermore, the stitching module is specifically used to stitch the feature map after the feature size is aligned and the refined vehicle feature information in the channel dimension through a large model feature stitching model preset in the vehicle detection network.

[0028] Furthermore, the system also includes: a training module;

[0029] The training module is used to detect the target through the loss function L = α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , the target detection loss function L, α1 and α2 are weight coefficients respectively.

[0030] The present invention provides a vehicle detection method and system based on a roadside parking scene, which adopts a target detection framework based on a region candidate frame. First, an input image is input into a backbone network for feature extraction to obtain a feature map, and then a vehicle candidate frame is obtained by using a region generation network. On the one hand, the vehicle candidate frame is sent to a candidate region alignment network to align the size of the input feature; on the other hand, the vehicle candidate frame is sent to a pre-trained vehicle large model to extract more refined vehicle features; finally, the features of the two aspects are spliced ​​in the channel dimension and sent to a vehicle detection and classification network to obtain a final output, which can achieve more accurate vehicle detection results and provide a more accurate judgment basis when performing parking management tasks by using roadside parking scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of a vehicle detection method based on a roadside parking scenario provided by the present invention;

[0032] Figure 2 It is a schematic diagram of a vehicle detection system based on a roadside parking scenario provided by the present invention. DETAILED DESCRIPTION

[0033] The device structure and implementation of the present invention are further described in detail below through the accompanying drawings and embodiments.

[0034] The present invention provides a vehicle detection method based on a roadside parking scenario, such as Figure 1 As shown, the specific steps include:

[0035] 101. Extract features of an image frame using a feature extraction model preset in a vehicle detection network, and perform feature aggregation on the feature information.

[0036] Specifically, the step of extracting features of image frames through a feature extraction model in a preset vehicle detection network and performing feature aggregation on the feature information includes: extracting features of image frames through a feature extraction model in a preset vehicle detection network according to the resolution size of the original video corresponding to the image frame, and fusing the feature information between different layers of the preset vehicle detection network.

[0037] For example, the input image size is the resolution size of the original video, which is defined as H*W*C, which represent the height, width and number of channels of the image respectively, and ResNet-18 is used as the backbone network; in addition, in order to more fully extract features for the subsequent learning of the vehicle detection network, feature aggregation is added on the basis of the backbone network. The function of the feature aggregation is to aggregate the high-level and low-level features extracted by the network, thereby improving the accuracy of subsequent target detection, because in the convolutional neural network, the high-level semantic features are close to the output end of the network but have a lower resolution, while the high-resolution features are close to the input end but have fewer semantic features. Therefore, the features in different layers of the network are fused to achieve the fusion between high-level and low-level features, thereby improving the subsequent detection accuracy. Specifically, the feature fusion operation includes but is not limited to the use of feature fusion methods such as FPN (FeaturePyramid Network) and ASPP (Atrous Spatial Pooling Pyramid).

[0038] For the embodiment of the present invention, before step 101, the following steps may be further included: using the target detection loss function L=α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , target detection loss function L, α1 and α2 are weight coefficients respectively.

[0039] The video surveillance camera installed on the roadside is used to capture the video image data of the parking scene, and the parking scene covers data under different surveillance camera installation scenes, different surveillance angles, and different weather conditions. The vehicle in the roadside parking scene image is annotated using a rectangular box annotation tool, which is specifically expressed as: O = [x, y, w, h, cls], where x, y, w, h represent the center point and width and height of the rectangular box respectively, and cls represents the category, which is vehicle.

[0040] Furthermore, a vehicle pre-trained large model VehicleMAE is defined; a general pre-trained large model, if directly applied to the vehicle single target detection task, may have problems of false detection and missed detection, and cannot achieve the expected effect. Therefore, a visual large model VehicleMAE that only targets vehicles is adopted to provide more refined features for vehicle detection; the VehicleMAE combines structural information such as the spatial structure from the vehicle contour information and the semantic structure from the information-rich high-level natural language description, and can effectively reconstruct the vehicle appearance.

[0041] Specifically, for VehicleMAE, where MAE stands for Masked Auto-Encoder, the masked auto-encoder performs random masking on the input image, learns through the auto-encoder, and outputs a complete reconstructed image. VehicleMAE uses a Transformer-based codec to achieve masked image reconstruction and constructs a spatial prior module to extract the contour information of the input image through an edge detector, and supervises the structural information of the reconstructed image through the contour information; in addition, a semantic prior module is constructed to encode the natural language description using the pre-trained visual language model CLIP; the cosine similarity loss is used to constrain the similarity between the CLIP visual embedding and the reconstructed features of the Transformer decoder; the cross-modal contrast learning loss is used to constrain the similarity distribution between the CLIP visual embedding, the Transformer decoder reconstructed features, and the CLIP semantic embedding; VehicleMAE forms an effective vehicle pre-training large model by utilizing a large amount of vehicle data and fully learning features such as the vehicle's structural appearance, providing more refined features for downstream visual tasks such as vehicle detection and vehicle attribute recognition.

[0042] 102. Extract candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in the vehicle detection network, and perform feature size alignment on the candidate frame position information by using a feature size alignment model preset in the vehicle detection network.

[0043] Specifically, the feature information after feature aggregation is transformed into a classification result and candidate box position information through a candidate area extraction model preset in a vehicle detection network; the candidate box position information is bilinearly interpolated through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers; and feature size alignment is performed according to the image values ​​at pixel points whose coordinates are floating point numbers.

[0044] It should be noted that the RPN refers to Region Proposal Network, which is a candidate region generation method. The candidate region contains positive objects and negative objects. The RPN input is the feature map obtained in the above steps. It first passes through a 3*3 convolution layer to transform the feature dimension, and then is sent to two 1*1 convolution layers respectively. One of the 1*1 convolution layers is used to connect the softmax classification function to score the positive objects and negative objects to obtain the classification results; the other 1*1 convolution layer is used for the parameters of the target detection box to obtain the accurate candidate box position.

[0045] Furthermore, ROI Align is used to align the feature sizes of the candidate regions obtained by the above RPN process. The ROI Align operation is proposed relative to ROIPooling to address the shortcomings of ROIPooling. ROIAlign cancels the quantization operation in ROIPooling and implements the bilinear interpolation method to obtain the image values ​​at the pixel points whose coordinates are floating-point numbers.

[0046] 103. Extract refined vehicle feature information from the candidate frame position information through a pre-trained large model in a preset vehicle detection network, and perform feature stitching on the feature map after feature size alignment and the refined vehicle feature information through a large model feature stitching model in the preset vehicle detection network.

[0047] Specifically, the step of performing feature stitching on the feature map after the feature size alignment and the refined vehicle feature information through the large model feature stitching model preset in the vehicle detection network includes: stitching the feature map after the feature size alignment and the refined vehicle feature information in the channel dimension through the large model feature stitching model preset in the vehicle detection network.

[0048] For example, the input of this module includes two parts. Input one is the feature map after feature size alignment after ROI Align in the above step; input two is the candidate areas obtained after passing through the RPN module. These candidate areas are cropped from the original input image according to the position coordinates to form the input data set of the vehicle large model. According to the above definition of the vehicle large model, the parameters of the vehicle large model are fixed, and after a multi-layer transformer structure, the encoded vehicle candidate frame features are obtained. In order to align the dimension with input one, the features are transformed in dimension using a linear mapping layer to obtain features with the same feature map size as input one; finally, input one and input two are spliced ​​in the channel dimension to realize large model feature splicing.

[0049] 104. Outputting a vehicle target detection result according to the feature map after the feature concatenation by using a vehicle classification and regression model preset in the vehicle detection network.

[0050] Specifically, the target classification and regression module, wherein the target classification is used to predict and output the category of the target, that is, the vehicle category; wherein the regression module is used to regress and output the rectangular box position of the vehicle target; wherein the target classification module and the target regression module are conventional convolution modules, which are composed of multiple convolution layers.

[0051] A vehicle detection method based on a roadside parking scenario provided by an embodiment of the present invention adopts a target detection framework based on a region candidate frame. First, an input image is input into a backbone network for feature extraction to obtain a feature map. Then, a vehicle candidate frame is obtained by using a region generation network. On the one hand, the vehicle candidate frame is sent to a candidate region alignment network to align the size of the input feature. On the other hand, the vehicle candidate frame is sent to a pre-trained vehicle large model to extract more refined vehicle features. Finally, the features of the two aspects are spliced ​​in the channel dimension and sent to a vehicle detection and classification network to obtain a final output. This can achieve more accurate vehicle detection results and provide a more accurate judgment basis when performing parking management tasks by using roadside parking scenarios.

[0052] As Figure 1 The specific implementation of the method shown in the figure, the embodiment of the present invention provides a vehicle detection system based on a roadside parking scenario, such as Figure 2 As shown, the system includes: an extraction module 21, which is used to extract features of image frames through a feature extraction model preset in a vehicle detection network, and perform feature aggregation on the feature information;

[0053] An alignment module 22 is used to extract candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in the vehicle detection network, and perform feature size alignment on the candidate frame position information by using a feature size alignment model preset in the vehicle detection network;

[0054] A stitching module 23 is used to extract refined vehicle feature information from the candidate frame position information through a pre-trained large model in a preset vehicle detection network, and perform feature stitching on the feature map after feature size alignment and the refined vehicle feature information through a large model feature stitching model in the preset vehicle detection network;

[0055] The target detection module 24 is used to output the vehicle target detection result according to the feature map after the feature splicing by using the vehicle classification and regression model preset in the vehicle detection network.

[0056] Furthermore, the extraction module 21 is specifically used to extract features of the image frame according to the resolution size of the original video corresponding to the image frame through a feature extraction model in a preset vehicle detection network, and to fuse feature information between different layers of the preset vehicle detection network.

[0057] Furthermore, the alignment module 22 is specifically used to perform feature dimension transformation on the feature information after feature aggregation through a candidate area extraction model preset in a vehicle detection network to obtain a classification result and candidate box position information; perform bilinear interpolation processing on the candidate box position information through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers; and perform feature size alignment according to the image values ​​at pixel points whose coordinates are floating point numbers.

[0058] Furthermore, the stitching module 23 is specifically used to stitch the feature map after the feature size is aligned and the refined vehicle feature information in the channel dimension by using a large model feature stitching model preset in the vehicle detection network.

[0059] Furthermore, the system further comprises: a training module 25;

[0060] The training module is used to detect the target through the loss function L = α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , the target detection loss function L, α1 and α2 are weight coefficients respectively.

[0061] A vehicle detection system based on a roadside parking scenario provided by an embodiment of the present invention adopts a target detection framework based on a region candidate frame. First, an input image is input into a backbone network for feature extraction to obtain a feature map. Then, a vehicle candidate frame is obtained by using a region generation network. On the one hand, the vehicle candidate frame is sent to a candidate region alignment network to align the size of the input feature. On the other hand, the vehicle candidate frame is sent to a pre-trained vehicle large model to extract more refined vehicle features. Finally, the features of the two aspects are spliced ​​in the channel dimension and sent to a vehicle detection and classification network to obtain a final output, which can achieve more accurate vehicle detection results and provide a more accurate judgment basis when performing parking management tasks by using roadside parking scenarios.

[0062] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0063] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0064] The disclosed embodiments are described above to enable any person skilled in the art to implement or use the present invention. Various modifications of these embodiments are obvious to those skilled in the art, and the general principles defined herein may also be applied to other embodiments without departing from the spirit and scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments given herein, but is consistent with the broadest scope of the principles and novel features disclosed in this application.

[0065] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

[0066] Those skilled in the art may also understand that the various illustrative logical blocks, units, and steps listed in the embodiments of the present invention may be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly demonstrate the interchangeability of hardware and software, the various illustrative components, units, and steps described above have generally described their functions. Whether such functions are implemented by hardware or software depends on the specific application and the design requirements of the entire system. Those skilled in the art may use various methods to implement the described functions for each specific application, but such implementation should not be understood as exceeding the scope of protection of the embodiments of the present invention.

[0067] The various illustrative logic blocks or units described in the embodiments of the present invention can be implemented or operated by a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of the above. The general-purpose processor can be a microprocessor, and optionally, the general-purpose processor can also be any conventional processor, controller, microcontroller or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration.

[0068] The steps of the method or algorithm described in the embodiments of the present invention can be directly embedded in hardware, a software module executed by a processor, or a combination of the two. The software module can be stored in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or other storage media of any form in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and the storage medium can be arranged in an ASIC, and the ASIC can be arranged in a user terminal. Optionally, the processor and the storage medium can also be arranged in different components in the user terminal.

[0069] In one or more exemplary designs, the above functions described in the embodiments of the present invention can be implemented in hardware, software, firmware or any combination of the three. If implemented in software, these functions can be stored on a computer-readable medium, or transmitted in the form of one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media that facilitate the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by any general or special computer. For example, such computer-readable media can include but are not limited to RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store program codes in the form of instructions or data structures and other forms that can be read by general or special computers, or general or special processors. In addition, any connection can be appropriately defined as a computer-readable medium, for example, if the software is transmitted from a website site, server or other remote resource through a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wirelessly, such as infrared, wireless and microwave, it is also included in the defined computer-readable medium. The disk and disc include compact disk, laser disk, optical disk, DVD, floppy disk and blue-ray disk. Disks usually copy data magnetically, while discs usually copy data optically with lasers. The above combination can also be included in computer readable media.

[0070] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A vehicle detection method based on a roadside parking scenario, characterized in that: The method comprises: Extracting features of the image frame by using a feature extraction model preset in the vehicle detection network, and performing feature aggregation on the feature information; Extracting candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model in a preset vehicle detection network, and aligning the feature size of the candidate frame position information by using a feature size alignment model in a preset vehicle detection network; Extracting refined vehicle feature information from the candidate frame position information by using a pre-trained large model in a preset vehicle detection network, and performing feature splicing on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature splicing model in the preset vehicle detection network; By presetting the vehicle classification and regression model in the vehicle detection network, the vehicle target detection result is output according to the feature map after the feature splicing.

2. A vehicle detection method based on roadside parking scenario according to claim 1, characterized in that: The step of extracting features of the image frame by using a feature extraction model preset in the vehicle detection network and performing feature aggregation on the feature information comprises: According to the resolution size of the original video corresponding to the image frame, the feature extraction model in the preset vehicle detection network is used to extract the features of the image frame, and the feature information between different layers of the preset vehicle detection network is fused.

3. A vehicle detection method based on roadside parking scenario according to claim 1 or 2, characterized in that: The step of extracting candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in the vehicle detection network, and aligning the feature size of the candidate frame position information by using a feature size alignment model preset in the vehicle detection network comprises: The feature information after the feature aggregation is transformed into a feature dimension by using a candidate region extraction model preset in a vehicle detection network to obtain a classification result and candidate frame position information; Performing bilinear interpolation processing on the candidate frame position information through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers; Feature size alignment is performed according to image values ​​at pixel points whose coordinates are floating point numbers.

4. The vehicle detection method based on roadside parking scenario according to claim 1, characterized in that: The step of performing feature stitching on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature stitching model preset in the vehicle detection network comprises: The feature map after the feature size alignment and the refined vehicle feature information are spliced ​​in the channel dimension through a large model feature splicing model preset in the vehicle detection network.

5. The vehicle detection method based on roadside parking scenario according to claim 1 is characterized in that: Before the step of extracting features of the image frame by using the feature extraction model preset in the vehicle detection network and performing feature aggregation on the feature information, the method further includes: Through the target detection loss function L = α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , target detection loss function L, α1 and α2 are weight coefficients respectively.

6. A vehicle detection system based on roadside parking scenarios, characterized in that: The system comprises: An extraction module, used to extract features of the image frame by using a feature extraction model preset in the vehicle detection network, and perform feature aggregation on the feature information; An alignment module, used to extract candidate frame position information from the feature information after feature aggregation by using a candidate region extraction model preset in a vehicle detection network, and perform feature size alignment on the candidate frame position information by using a feature size alignment model preset in the vehicle detection network; A splicing module, used to extract refined vehicle feature information from the candidate frame position information by using a pre-trained large model in a preset vehicle detection network, and to perform feature splicing on the feature map after feature size alignment and the refined vehicle feature information by using a large model feature splicing model in the preset vehicle detection network; The target detection module is used to output the vehicle target detection result according to the feature map after the feature splicing by presetting the vehicle classification and regression model in the vehicle detection network.

7. A vehicle detection system based on roadside parking scenarios according to claim 6, characterized in that: The extraction module is specifically used to extract features of the image frame according to the resolution size of the original video corresponding to the image frame through a feature extraction model in a preset vehicle detection network, and to fuse feature information between different layers of the preset vehicle detection network.

8. The vehicle detection system based on roadside parking scenario according to claim 6, characterized in that: The alignment module is specifically used to perform feature dimension transformation on the feature information after feature aggregation through a candidate area extraction model preset in a vehicle detection network to obtain a classification result and candidate box position information; perform bilinear interpolation processing on the candidate box position information through a feature size alignment model preset in a vehicle detection network to obtain image values ​​at pixel points whose coordinates are floating point numbers; and perform feature size alignment according to the image values ​​at the pixel points whose coordinates are floating point numbers.

9. The vehicle detection system based on roadside parking scenario according to claim 6, characterized in that: The splicing module is specifically used to splice the feature map after the feature size is aligned and the refined vehicle feature information in the channel dimension through a large model feature splicing model preset in the vehicle detection network.

10. The vehicle detection system based on roadside parking scenario according to claim 6, characterized in that: The system also includes: a training module; The training module is used to detect the target through the loss function L = α1L bbox +α2L cls The preset vehicle detection network is trained and modified, wherein the position regression loss function is L bbox , the classification loss function is L cls , the target detection loss function L, α1 and α2 are weight coefficients respectively.

Images