Method and system for in-vehicle object tracking based on position in previous frame
By utilizing pre-trained convolutional network models and feature map mapping in in-vehicle human detection, the computational complexity of in-vehicle human detection is reduced, achieving efficient target tracking and recognition, and solving the problem of high computational cost in existing technologies.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MOMENTA (SUZHOU) TECHNOLOGY CO LTD
- Filing Date
- 2021-12-31
- Publication Date
- 2026-07-03
AI Technical Summary
Existing methods for detecting human bodies inside vehicles involve large amounts of computation and are complex in the process of target recognition and target tracking between consecutive frames, especially the IoU matching algorithm, which has high computational complexity.
A pre-trained convolutional network model is used to detect objects in the first frame of the image captured by the vehicle camera, and the bounding boxes in the first frame are determined. The number of semantic hierarchical model channels of the convolutional network model is reduced by feature map mapping and classification regression. The target location information of the previous frame is used to locate and track the target in the current frame.
This reduces the computational load and simplifies the target detection and tracking process, while ensuring the accuracy and efficiency of target recognition.
Smart Images
Figure CN116433716B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of autonomous driving technology, and in particular to a method and system for tracking in-vehicle objects based on their position in the previous frame. Background Technology
[0002] During vehicle operation, to fully understand the situation of people inside the vehicle and prevent safety issues such as driver or passenger errors or children being left behind, it is necessary to detect the presence of people inside the vehicle. Existing methods for detecting people inside vehicles employ general object detection techniques. After acquiring images of the vehicle interior using an in-vehicle camera, a detection model identifies and fuses these images to detect potential human bodies. Furthermore, object detection is a continuous process; after identifying objects in each frame, object tracking between consecutive frames is also required. Existing object tracking algorithms, such as the IoU (Intersection of Union) matching algorithm, involve computationally intensive and complex processes, including performing object detection in the current frame followed by matching with the target in the previous frame. Summary of the Invention
[0003] To address the issues of high computational load and complexity in existing technologies for target recognition and tracking of the same target across multiple consecutive frames, this application proposes a method and system for tracking in-vehicle objects based on their position in the previous frame.
[0004] One technical solution of this application provides a method for tracking in-vehicle objects based on their position in the previous frame, comprising: performing object detection on a first frame image captured by an in-vehicle camera using a pre-trained convolutional network model to determine the first bounding box corresponding to the target in the first frame image; extracting features from the current frame image using the convolutional network model to obtain a feature map corresponding to the current frame image; mapping the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determining the current center point of the current bounding box corresponding to the target in the current frame image on the feature map; and performing classification and regression processing on the feature map using the convolutional network model, and extracting the classification and regression result corresponding to the current center point to obtain the current bounding box.
[0005] Optionally, the number of model channels in each semantic layer of the Backbone layer in the convolutional network model can be reduced, and the model channels can be used to extract features from the current frame image to obtain the corresponding feature map.
[0006] Optionally, the previous center point of the previous bounding box corresponding to the target in the previous frame image is mapped onto the feature map, and the current center point of the current bounding box corresponding to the target in the current frame image is determined on the feature map. This includes: determining the previous center point corresponding to the previous bounding box and obtaining the camera coordinate system coordinates of the previous center point as the coordinates of the previous center point; adjusting the coordinates of the previous center point according to the reduction ratio between the current frame image and the feature map to obtain the coordinates of the current center point, and then obtaining the current center point corresponding to the coordinates of the current center point on the feature map.
[0007] Optionally, a convolutional network model is used to perform classification and regression processing on the feature map, and the classification and regression results corresponding to the current center point are extracted to obtain the current bounding box. This includes: using the convolutional network model to classify the feature points according to the target type to which the feature points constituting the feature map belong, and obtaining multiple feature point classification results; extracting the feature point classification results of the target corresponding to the current center point from the multiple feature point classification results; and determining the current bounding box of the target based on the feature point classification results of the target.
[0008] Optionally, the pre-training of the convolutional network model includes: obtaining a standard dataset containing multiple frames of original images; translating and / or rotating the multiple frames of original images to obtain corresponding multi-frame transformed images, forming a tracking dataset; and training the original convolutional network model using the standard dataset and the tracking dataset to obtain the convolutional network model.
[0009] In one technical solution of this application, a system for tracking in-vehicle objects based on their position in the previous frame is provided, comprising: a first frame image target determination module, which uses a pre-trained convolutional network model to perform target detection on the first frame image captured by an in-vehicle camera and determines the first bounding box corresponding to the target in the first frame image; a current frame image detection module, which uses a convolutional network model to extract features from the current frame image to obtain a feature map corresponding to the current frame image, and maps the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determines the current center point of the current bounding box corresponding to the target in the current frame image on the feature map; and a current frame image target determination module, which uses a convolutional network model to perform classification and regression processing on the feature map, and extracts the classification and regression result corresponding to the current center point to obtain the current bounding box.
[0010] The beneficial effects of this application are: This application uses the target detection result in the previous frame to perform target localization when detecting the target in the current frame, thereby performing target detection at the corresponding position, which can reduce the amount of computation in the target detection process. At the same time, the target determined in the current frame is the same target as the target determined in the previous frame, directly completing the target tracking, saving the complex computation of the tracking algorithm, and reducing the complexity of the target detection and tracking process. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 This is a flowchart illustrating one implementation of the method for tracking in-vehicle objects based on the position in the previous frame, as described in this application.
[0013] Figure 2 This is a schematic diagram of the basic structure of a convolutional network model;
[0014] Figure 3 This is a schematic diagram of one implementation of the system for tracking in-vehicle objects based on the position in the previous frame, as described in this application.
[0015] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a product or device comprising a series of steps or units is not necessarily limited to those units explicitly listed, but may include other units not explicitly listed or inherent to such products or devices.
[0018] During vehicle operation, to fully understand the situation of people inside the vehicle and prevent safety issues such as driver or passenger errors or children being left behind, it is necessary to detect the presence of people inside the vehicle. Existing methods for detecting people inside vehicles employ general object detection techniques. After acquiring images of the vehicle interior using an in-vehicle camera, a detection model identifies and fuses these images to detect potential human bodies. Furthermore, object detection is a continuous process; after identifying objects in each frame, object tracking between consecutive frames is also required. Existing object tracking algorithms, such as the IoU (Intersection of Union) matching algorithm, perform object detection in the current frame followed by matching with the object in the previous frame, resulting in high computational complexity and low accuracy.
[0019] Therefore, when performing target detection, this application uses the position of the target detection result of the previous frame as a basis for the target detection of the current frame. The target of the current frame determined by the target detection result of the previous frame has a strong correlation with the target, which is the same target. After reducing the complexity and computational load of the target detection process in the current frame, the tracking process of the same target between the targets detected in different frames is completed at the same time, omitting complex tracking algorithms and reducing the computational load.
[0020] To address this, this application proposes a method and system for in-vehicle object tracking based on the position in the previous frame. The method includes: using a pre-trained convolutional network model to perform object detection on the first frame image captured by the vehicle camera, and determining the first bounding box corresponding to the target in the first frame image; using the convolutional network model to extract features from the current frame image, and obtaining the feature map corresponding to the current frame image; mapping the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determining the current center point of the current bounding box corresponding to the target in the current frame image on the feature map; and using the convolutional network model to perform classification and regression processing on the feature map, and extracting the classification and regression result corresponding to the current center point to obtain the current bounding box.
[0021] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.
[0022] Figure 1 This paper illustrates one implementation of the method for tracking in-vehicle objects based on their position in the previous frame.
[0023] exist Figure 1In the embodiment shown, the method for tracking in-vehicle objects based on the position in the previous frame of this application includes process S101, which uses a pre-trained convolutional network model to perform target detection on the first frame image captured by the vehicle camera, and determines the first bounding box corresponding to the target in the first frame image.
[0024] In this embodiment, the method of this application checks and tracks targets in the current frame image based on the position of the target in the previous frame. Therefore, when detecting targets in the first frame image, conventional target detection methods are used. Specifically, a pre-trained convolutional network model can be used to detect targets in the first frame image of the vehicle interior captured by the vehicle-mounted camera, ultimately obtaining multiple first detection boxes corresponding to targets inside the vehicle in the first frame image. Targets inside the vehicle include passengers, bags, toys, and various other possible items left behind. The selection of targets inside the vehicle can be chosen according to specific detection requirements, and the corresponding process can be trained during model training to complete the detection of the corresponding targets.
[0025] Specifically, in-vehicle cameras are typically installed at the upper front of the vehicle to effectively detect information from all seats. The camera's shooting angle should minimize obstruction between people. The installation location of the in-vehicle cameras and the resolution of the images can be reasonably set according to the actual vehicle conditions; this application does not impose specific restrictions.
[0026] Figure 2 A schematic diagram of the basic structure of a convolutional network model is shown.
[0027] like Figure 2 As shown, a basic convolutional network model generally includes a backbone layer, a feature pyramid, and a detection head. The backbone layer comprises multiple semantic layers, extracting higher-level semantic features from the in-vehicle image through layer-by-layer convolution, from layer C3 to layer C4, and finally to layer C5. The extracted speech feature map is then fused with the corresponding part of the feature pyramid, and the final result is input into the corresponding detection head. In the object recognition task, the detection head primarily performs object classification and regression processing, ultimately identifying the target in the image and obtaining its corresponding bounding box to complete object detection.
[0028] exist Figure 1 In the embodiment shown, the method for tracking in-vehicle objects based on the position in the previous frame of this application includes process S102, which uses a convolutional network model to extract features from the current frame image to obtain the feature map corresponding to the current frame image.
[0029] Optionally, a convolutional network model can be used to extract features from the current frame image to obtain the corresponding feature map. This includes reducing the number of model channels in each semantic layer of the Backbone layer of the convolutional network model and using a lower number of model channels to extract features from the current frame image to obtain the corresponding feature map.
[0030] In this optional embodiment, the detection of people inside the vehicle using in-vehicle images does not require excessively precise image information to complete the detection task. Furthermore, the more model channels analyzed in the image, the greater the corresponding data processing volume, increasing the model's data processing burden. Here, a model channel corresponds to a specific image feature for detection, such as texture or color. Therefore, when performing in-vehicle human detection, only the model channels corresponding to the features useful for human detection need to be retained, accurately completing the detection while reducing the data processing volume. Thus, unlike existing technologies that use standard convolutional network models for in-vehicle human detection, this application reduces the number of model channels in each semantic layer of the Backbone layer of the convolutional network model, utilizing a lower number of model channels to complete feature extraction from in-vehicle images.
[0031] Specifically, in the implementation of the seat-based in-vehicle human detection method of this application, the number of model channels in each semantic layer C3-C5 of the Backbone base layer can be halved. For example, the number of channels in 32 convolutional layers can be halved to 16. By reducing the number of model channels, the amount of data processing for the convolutional network model is greatly reduced, and the data processing speed is improved.
[0032] exist Figure 1 In the embodiment shown, the method for tracking in-vehicle objects based on the position in the previous frame of this application includes the process S103, which maps the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determines the current center point of the current bounding box corresponding to the target in the current frame image on the feature map.
[0033] In this implementation, after extracting features from the current frame image using the Backbone layer of the convolutional network model to obtain the feature map, the center point of the previous bounding box of the vehicle interior target recognition result in the previous frame image is projected onto the feature map of the current frame image. This ultimately determines the current center point of the target's current bounding box in the current frame image, corresponding to the previous center point of the previous bounding box. During feature map extraction in the Backbone layer of the convolutional network model, image size is compressed layer by layer; therefore, when mapping and projecting the center point, the current center point is determined based on the image compression ratio.
[0034] Optionally, the previous center point of the previous bounding box corresponding to the target in the previous frame image is mapped onto the feature map, and the current center point of the current bounding box corresponding to the target in the current frame image is determined on the feature map. This includes: determining the previous center point corresponding to the previous bounding box and obtaining the camera coordinate system coordinates of the previous center point as the coordinates of the previous center point; adjusting the coordinates of the previous center point according to the reduction ratio between the current frame image and the feature map to obtain the coordinates of the current center point, and then obtaining the current center point corresponding to the coordinates of the current center point on the feature map.
[0035] In this optional embodiment, after determining the bounding boxes of the targets inside the vehicle in the previous frame image (e.g., rectangular bounding boxes), the center points of each bounding box are then determined. This yields the coordinates of the previous center point within the camera coordinate system of the previous frame image. Therefore, during feature map extraction, the image is scaled down proportionally, and the original coordinates of the previous center point are reduced to the corresponding values. Based on this relationship, the current center point coordinates corresponding to the previous center point are determined in the feature map of the current frame image. Once the current center point coordinates are determined, the center point corresponding to these coordinates is the current center point of the target inside the vehicle on the feature map of the current frame.
[0036] Specifically, if we take the top-left vertex of the image as the origin, with the positive X-axis pointing to the right and the positive Y-axis pointing downwards, the coordinates of the previous center point of the target inside the car in the previous frame are (8, 8). After feature extraction through the Backbone layer of the convolutional network model, for example, with each of the three semantic layers C3-C5 having a compression ratio of 2, the final image compression is 8. Therefore, the corresponding coordinates of the previous center point on the feature map of the previous frame are (1, 1). Thus, the current center point coordinates on the feature map of the current frame are determined to be (1, 1). This point is the current center point of the target's bounding box in the current frame.
[0037] exist Figure 1 In the embodiment shown, the method for tracking in-vehicle objects based on the position in the previous frame of this application includes process S104, which uses a convolutional network model to perform classification and regression processing on the feature map, and extracts the classification and regression result corresponding to the current center point to obtain the current bounding box.
[0038] In this embodiment, after obtaining the feature map corresponding to the current frame image in the above process, a convolutional network model is used to perform target classification and regression processing on each point on the feature map to obtain the classification and regression results for each point. Simultaneously, based on the current center point coordinates, the classification and regression results corresponding to the current center point are extracted from the classification results of each point, which serve as the bounding boxes for target detection in the current frame image, completing target detection and target tracking, thus identifying the same target across different frames. Compared to existing target detection methods, which require target fusion and matching after completing target classification and regression processing at each position on the feature map to determine the target, this application directly uses the classification result of the current center point as the target recognition result, making it simpler and more efficient. Therefore, when performing target detection and tracking in the current frame image based on the target detection position in the previous frame, the computational load is greatly reduced, while the accuracy of target recognition is well guaranteed.
[0039] Optionally, a convolutional network model is used to perform classification and regression processing on the feature map, and the classification and regression results corresponding to the current center point are extracted to obtain the current bounding box. This includes: using the convolutional network model to classify the feature points according to the target type to which the feature points constituting the feature map belong, and obtaining multiple feature point classification results; extracting the feature point classification results of the target corresponding to the current center point from the multiple feature point classification results; and determining the current bounding box of the target based on the feature point classification results of the target.
[0040] In this optional embodiment, a convolutional network model is used to classify each feature point in the feature map. Based on the attributes of different targets corresponding to each feature point, they are classified into different targets, such as a backpack or a mobile phone. Then, based on the classification results of the feature points corresponding to the current center point, corresponding regression processing is performed to obtain the current bounding box of the target corresponding to the current center point. By linking the previous center point in the previous frame image with the current center point in the current frame image, the same target in different frames is connected, omitting the problem of tracking the same target in traditional methods and reducing computational load.
[0041] Optionally, the method for tracking in-vehicle objects based on the position in the previous frame in this application further includes: pre-training of a convolutional network model, including: obtaining a standard dataset containing multiple frames of original images; translating and / or rotating the multiple frames of original images respectively to obtain corresponding multi-frame transformed images, forming a tracking dataset; and training the original convolutional network model using the standard dataset and the tracking dataset to obtain a convolutional network model.
[0042] In this embodiment, the method for tracking in-vehicle objects based on their position in the previous frame utilizes the characteristic that the target's position changes very little across consecutive frames, almost remaining identical. By using the target's position in the previous frame, the position of the target in the current frame can be determined, greatly ensuring the accuracy of target detection. Furthermore, the pre-trained convolutional network model in this application is also trained during the training phase to guarantee the accuracy of target detection.
[0043] First, during model training, a standard dataset containing multiple frames of original images is acquired. This labeled dataset serves as the preceding frame in the actual object detection inference process. Then, positional transformations are performed on the images in the standard dataset to obtain the corresponding tracking dataset. These transformations include translation and rotation, simulating real-world object movement and pose changes within a vehicle. The convolutional network model is then trained using both the labeled and tracking datasets. Utilizing the concept of receptive fields, the model can accurately predict the position of the detected object in the tracking dataset based on the image from the standard dataset. A loss function is used to gradually narrow the gap between the predicted and actual positions of the object in the tracking dataset, improving model accuracy. Ultimately, by inputting two consecutive frames into the model, the model can directly predict the target position in the next frame based on the target position identified in the previous frame. This reduces the complexity and computational load of object detection in the next frame and eliminates the problem of tracking the same target across multiple frames.
[0044] The method for tracking in-vehicle objects based on the position in the previous frame of this application utilizes the target detection results in the previous frame to locate a target during target detection in the current frame, thereby performing target detection at the corresponding position. This reduces the computational load of the target detection process. At the same time, the target determined in the current frame is the same target determined in the previous frame, directly completing target tracking, saving the complex computational load of the tracking algorithm, and reducing the complexity of the target detection and tracking process.
[0045] Figure 3 This paper illustrates one implementation of the system for tracking in-vehicle objects based on their position in the previous frame.
[0046] exist Figure 3In the illustrated embodiment, the in-vehicle object tracking system based on the position in the previous frame of this application includes a first frame image target determination module 301, which uses a pre-trained convolutional network model to perform target detection on the first frame image captured by the vehicle camera and determines the first bounding box corresponding to the target in the first frame image; a current frame image detection module 302, which uses the convolutional network model to extract features from the current frame image to obtain a feature map corresponding to the current frame image, and maps the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determines the current center point of the current bounding box corresponding to the target in the current frame image on the feature map; and a current frame image target determination module 303, which uses the convolutional network model to perform classification and regression processing on the feature map, and extracts the classification and regression result corresponding to the current center point to obtain the current bounding box.
[0047] Optionally, in the current frame image detection module, the previous center point is determined based on the previous bounding box, and the camera coordinates of the previous center point are obtained as the coordinates of the previous center point; the coordinates of the previous center point are adjusted according to the reduction ratio between the current frame image and the feature map to obtain the coordinates of the current center point, and then the current center point corresponding to the coordinates of the current center point is obtained on the feature map.
[0048] Optionally, in the target determination module of the current frame image, a convolutional network model is used to classify the feature points according to the target type to which the feature points constituting the feature map belong, and multiple feature point classification results are obtained; among the multiple feature point classification results, the feature point classification result of the target corresponding to the current center point is extracted; and the current bounding box of the target is determined according to the feature point classification result of the target.
[0049] The system for tracking in-vehicle objects based on the position in the previous frame of this application utilizes the target detection results in the previous frame to locate a target during target detection in the current frame, thereby performing target detection at the corresponding position. This reduces the computational load of the target detection process. At the same time, the target determined in the current frame is the same target determined in the previous frame, directly completing target tracking, saving the complex computational load of the tracking algorithm, and reducing the complexity of the target detection and tracking process.
[0050] In one specific embodiment of this application, a computer-readable storage medium stores computer instructions, wherein the computer instructions are operated to perform the in-vehicle object tracking method based on position in a previous frame described in any embodiment. The storage medium may be located directly in hardware, in a software module executed by a processor, or in a combination of both.
[0051] Software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in this art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium.
[0052] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor can be a microprocessor, but alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors incorporating a DSP core, or any other such configuration. Alternatively, the storage medium can be integrated with the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in the user terminal. Alternatively, the processor and storage medium can reside as discrete components in the user terminal.
[0053] In one specific embodiment of this application, a computer device includes a processor and a memory, the memory storing computer instructions, wherein the processor operates the computer instructions to perform the in-vehicle object tracking method based on position in a previous frame as described in any embodiment.
[0054] In the embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0055] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0056] The above are merely embodiments of this application and do not limit the scope of this patent application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.
Claims
1. A method for tracking in-vehicle objects based on their position in the previous frame, characterized in that, include: The first frame image captured by the vehicle camera is used to detect targets using a pre-trained convolutional network model, and the first bounding box corresponding to the target in the first frame image is determined. The convolutional network model is used to extract features from the current frame image to obtain the feature map corresponding to the current frame image; Mapping the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determining the current center point of the current bounding box corresponding to the target in the current frame image on the feature map, wherein the step of mapping the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map, and determining the current center point of the current bounding box corresponding to the target in the current frame image on the feature map includes: The previous center point is determined based on the previous annotation box, and the camera coordinate system coordinates of the previous center point are obtained as the coordinates of the previous center point. Based on the reduction ratio between the current frame image and the feature map, the coordinates of the previous center point are adjusted to obtain the coordinates of the current center point, and then the current center point corresponding to the current center point coordinates is obtained on the feature map; and The convolutional network model is used to perform classification and regression processing on the feature map, and the classification and regression results corresponding to the current center point are extracted to obtain the current bounding box.
2. The method for tracking in-vehicle objects based on position in the previous frame according to claim 1, characterized in that, The step of extracting features from the current frame image using the convolutional network model to obtain the feature map corresponding to the current frame image includes: The number of data channels in each semantic layer of the Backbone layer in the convolutional network model is reduced, and the data channels are used to extract features from the current frame image to obtain the corresponding feature map.
3. The method for tracking in-vehicle objects based on their position in the previous frame according to claim 1, characterized in that, The step of using the convolutional network model to perform classification and regression processing on the feature map, and extracting the classification and regression result corresponding to the current center point to obtain the current bounding box includes: The convolutional network model is used to classify the feature points according to the target type to which the feature points constituting the feature map belong, resulting in multiple feature point classification results; Extract the feature point classification result of the target corresponding to the current center point from the multiple feature point classification results; Based on the feature point classification results of the target, the current bounding box of the target is determined.
4. The method for tracking in-vehicle objects based on their position in the previous frame according to claim 1, characterized in that, The pre-training of the convolutional network model includes: Obtain a standard dataset containing multiple frames of original images; The original images of the multiple frames are translated and / or rotated respectively to obtain corresponding transformed images of the multiple frames, which are then used to form a tracking dataset. The original convolutional network model is trained using the standard dataset and the tracking dataset to obtain the convolutional network model.
5. A system for tracking in-vehicle objects based on their position in the previous frame, characterized in that, include: The first frame image target determination module uses a pre-trained convolutional network model to perform target detection on the first frame image captured by the vehicle camera, and determines the first bounding box corresponding to the target in the first frame image; The current frame image detection module uses the convolutional network model to extract features from the current frame image, obtains a feature map corresponding to the current frame image, and maps the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map. It then determines the current center point of the current bounding box corresponding to the target in the current frame image on the feature map. Specifically, mapping the previous center point of the previous bounding box corresponding to the target in the previous frame image to the feature map and determining the current center point of the current bounding box corresponding to the target in the current frame image on the feature map includes: determining the previous center point corresponding to the previous bounding box, and obtaining the camera coordinate system coordinates of the previous center point as the coordinates of the previous center point. Based on the reduction ratio between the current frame image and the feature map, the coordinates of the previous center point are adjusted to obtain the coordinates of the current center point, and then the current center point corresponding to the current center point coordinates is obtained on the feature map; and The current frame image target determination module uses the convolutional network model to perform classification and regression processing on the feature map, and extracts the classification and regression result corresponding to the current center point to obtain the current bounding box.
6. The system for tracking in-vehicle objects based on position in the previous frame according to claim 5, characterized in that, In the current frame image target determination module, the convolutional network model is used to classify the feature points according to the target type to which the feature points constituting the feature map belong, and multiple feature point classification results are obtained. Among the multiple feature point classification results, the feature point classification result of the target corresponding to the current center point is extracted, and the current bounding box of the target is determined according to the feature point classification result of the target.
7. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions that are operated to perform the in-vehicle object tracking method based on the position in the previous frame as described in any one of claims 1-4.
8. A computer device comprising a processor and a memory, the memory storing computer instructions, wherein: The processor operates computer instructions to execute the method for in-vehicle object tracking based on position in the previous frame as described in any one of claims 1-4.