Vehicle tracking method and system integrating camera and radar

Abstract

The present invention discloses a vehicle tracking method and system that integrates cameras and radars, belonging to the field of intelligent vehicles. The method includes: deploying multiple vehicle tracking components at preset intervals in a first driving area; sensing real-time light intensity and determining whether it is greater than or equal to a light intensity threshold; when the real-time light intensity is greater than or equal to the light intensity threshold, starting the camera to perform image acquisition, obtaining a first image acquisition result, and generating a first vehicle identification result; when the real-time light intensity is less than the light intensity threshold, starting the tracking radar to perform image acquisition, obtaining a second image acquisition result, and generating a second vehicle identification result; and splicing multiple vehicle identification results to generate a vehicle tracking result. This application solves the technical problem in the prior art of over-reliance on cameras in complex scenarios, resulting in unstable target tracking, and achieves the technical effect of intelligently selecting sensors under different lighting conditions and improving the robustness and accuracy of vehicle tracking.

Description

Technical Field

[0001] The present invention relates to the field of intelligent vehicles, and in particular to a vehicle tracking method and system integrating cameras and radars. Background Art

[0002] With the development of autonomous driving and intelligent transportation systems, the demand for vehicle tracking technology is gradually increasing. Traditional vehicle tracking systems usually rely on cameras for target detection and tracking. However, this method is easily limited in complex road environments, such as poor lighting conditions, resulting in low target tracking accuracy and stability. Summary of the Invention

[0003] This application provides a vehicle tracking method and system that integrates cameras and radars, aiming to solve the technical problem in the prior art of unstable target tracking caused by excessive reliance on cameras in complex scenarios.

[0004] In view of the above problems, the present application provides a vehicle tracking method and system that integrates cameras and radars.

[0005] The first aspect disclosed in the present application provides a vehicle tracking method that integrates a camera and a radar, the method comprising: deploying multiple vehicle tracking components at preset intervals in a first driving area, wherein any one of the multiple vehicle tracking components comprises a tracking radar, a camera, and a light sensor; sensing real-time light intensity through the light sensor to determine whether it is greater than or equal to a light intensity threshold; when the real-time light intensity is greater than or equal to the light intensity threshold, starting the camera to perform image acquisition, obtaining a first image acquisition result, activating a first target detection channel to perform target recognition, and generating a first vehicle recognition result; when the real-time light intensity is less than the light intensity threshold, starting the tracking radar to perform image acquisition, obtaining a second image acquisition result, activating a second target detection channel to perform target recognition, and generating a second vehicle recognition result; and splicing multiple vehicle recognition results of the multiple vehicle tracking components to generate a vehicle tracking result, wherein the first vehicle recognition result and / or the second vehicle recognition result belong to the multiple vehicle recognition results.

[0006] Another aspect disclosed in the present application provides a vehicle tracking system that integrates cameras and radars, the system comprising: a tracking component deployment unit, configured to deploy multiple vehicle tracking components at preset intervals in a first driving area, wherein any one of the multiple vehicle tracking components comprises a tracking radar, a camera, and a light sensor; a light intensity judgment unit, configured to sense real-time light intensity through the light sensor and determine whether it is greater than or equal to a light intensity threshold; a first vehicle identification unit, configured to start the camera for image acquisition when the real-time light intensity is greater than or equal to the light intensity threshold, obtain a first image acquisition result, activate a first target detection channel for target recognition, and generate a first vehicle recognition result; a second vehicle identification unit, configured to start the tracking radar for image acquisition when the real-time light intensity is less than the light intensity threshold, obtain a second image acquisition result, activate a second target detection channel for target recognition, and generate a second vehicle recognition result; and a vehicle tracking result unit, configured to splice multiple vehicle recognition results of the multiple vehicle tracking components to generate a vehicle tracking result, wherein the first vehicle recognition result and / or the second vehicle recognition result belong to the multiple vehicle recognition results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Due to the deployment of multiple vehicle tracking components at preset intervals in the first driving area, multiple sensor components are deployed around the vehicle, providing a hardware foundation for vehicle tracking; the real-time light intensity is sensed by the light sensor to determine whether it is greater than or equal to the light intensity threshold, so as to select the appropriate sensor under different lighting conditions; when the real-time light intensity is greater than or equal to the light intensity threshold, the camera is started to collect images, obtain the first image collection result, activate the first target detection channel for target recognition, and generate the first vehicle recognition result, ensuring that the camera is used for target recognition under good lighting conditions; when the real-time light intensity is less than the light intensity threshold, The tracking radar is started to perform image acquisition, a second image acquisition result is obtained, a second target detection channel is activated for target recognition, and a second vehicle recognition result is generated, ensuring that under conditions of poor lighting, the radar is switched to obtain a more stable target recognition result; multiple vehicle recognition results of multiple vehicle tracking components are spliced ​​to generate a vehicle tracking result, and the recognition results of different sensors are comprehensively utilized to improve the robustness and accuracy of vehicle tracking. This technical solution solves the technical problem in the existing technology of unstable target tracking caused by excessive reliance on cameras in complex scenarios, and achieves the technical effect of intelligently selecting sensors under different lighting conditions and improving the robustness and accuracy of vehicle tracking.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of a vehicle tracking method integrating a camera and a radar is provided for an embodiment of the present application;

[0011] Figure 2 A schematic diagram of a process for generating vehicle tracking results in a vehicle tracking method integrating a camera and a radar is provided for an embodiment of the present application;

[0012] Figure 3 A structural diagram of a vehicle tracking system integrating a camera and a radar is provided for an embodiment of the present application.

[0013] Description of the accompanying drawings: tracking component deployment unit 11, light intensity judgment unit 12, first vehicle identification unit 13, second vehicle identification unit 14, vehicle tracking result unit 15. DETAILED DESCRIPTION

[0014] The overall idea of ​​the technical solution provided by this application is as follows:

[0015] The embodiments of the present application provide a vehicle tracking method and system that integrates cameras and radars. Specifically, multiple vehicle tracking components are deployed in a first driving area, each component including a tracking radar, a camera, and a light sensor. The light sensor intelligently senses the real-time light intensity and determines whether to start the camera or tracking radar based on the real-time light intensity, thereby selecting the sensor that best suits the lighting conditions. When the lighting conditions are good, the camera is used for image acquisition and target recognition. When the lighting conditions are poor, the radar is intelligently switched to obtain more stable tracking results, alleviating the problems that traditional cameras are prone to in such situations. This improves the performance and robustness of the vehicle tracking system under different lighting conditions, providing important support for autonomous driving and intelligent transportation systems.

[0016] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically introduced in conjunction with the drawings in the specification.

[0017] Example 1

[0018] like Figure 1 As shown, an embodiment of the present application provides a vehicle tracking method that integrates a camera and a radar, the method comprising:

[0019] Deploying a plurality of vehicle tracking components at preset intervals in a first driving area, wherein any one of the plurality of vehicle tracking components includes a tracking radar, a camera, and a light sensor;

[0020] In an embodiment of the present application, the first driving area is a specific area in a road or traffic scene, usually in the direction of the vehicle's advance, such as a section of road or an intersection. The preset interval is a predetermined distance used to determine the placement position of each vehicle tracking component, and the interval is adjusted according to the specific road and the needs of the vehicle tracking system. The vehicle tracking component is a device equipped with a variety of sensors and devices, which is used to monitor and track surrounding vehicles to achieve vehicle tracking and provide vehicle location and dynamic information. The vehicle tracking component includes a tracking radar, a camera and a light sensor, wherein the tracking radar is used to measure the distance, speed and position of the target vehicle by emitting radio waves and receiving reflected signals; the camera is used to capture images and videos on the road; the light sensor is used to sense and measure lighting conditions, and can detect the light intensity in the environment so as to select appropriate sensors and data acquisition methods according to changes in lighting conditions.

[0021] Specifically, multiple vehicle tracking units are installed within a first driving zone on the road, spaced at predetermined intervals. Each unit is equipped with a tracking radar, a camera, and a light sensor. These sensors work together to monitor and track the location and behavior of surrounding vehicles. The tracking radar provides accurate distance and speed information, the camera captures images for target recognition, and the light sensor helps select the sensor that best suits the current lighting conditions. This comprehensive arrangement effectively provides reliable vehicle tracking performance under varying lighting conditions.

[0022] The light sensor senses the real-time light intensity and determines whether it is greater than or equal to a light intensity threshold;

[0023] In this embodiment of the present application, real-time light intensity refers to the intensity or brightness level of light currently sensed by the light sensor, i.e., the value measured at a specific point in time. The real-time light intensity value will vary depending on the lighting conditions in the environment. The light intensity threshold is used to determine when to switch or select a different sensor or operating mode.

[0024] Specifically, the light sensor in each vehicle tracking component is activated to sense the current ambient light intensity, generating real-time light intensity, which reflects the intensity of light, which may come from sunlight, streetlights, or other light sources. This real-time light intensity is then compared with a light intensity threshold to determine the relationship between the real-time light intensity and the light intensity threshold. This intelligently selects the appropriate sensor for vehicle tracking based on the ambient lighting conditions, ensuring optimal vehicle tracking performance under varying lighting conditions and improving vehicle tracking stability and adaptability.

[0025] When the real-time light intensity is greater than or equal to the light intensity threshold, starting the camera to perform image acquisition, obtaining a first image acquisition result, activating a first target detection channel to perform target recognition, and generating a first vehicle recognition result;

[0026] Furthermore, this step specifically includes:

[0027] Collecting a first training data set, wherein the first training data set includes k sets of one-to-one corresponding first training image sets and first vehicle identification image sets;

[0028] Perform k random samplings with replacement on k groups of one-to-one corresponding first training image sets and first vehicle identification image sets to generate a first reconstructed training data set;

[0029] Retrieving the first reconstructed training data set, training a convolutional neural network, and generating a first target detection node;

[0030] Repeat M times to obtain the second target detection node until the Mth target detection node;

[0031] Merging the first target detection node, the second target detection node, up to the M-th target detection node as parallel nodes to generate the first target detection channel, wherein the output value of the first target detection channel is the highest frequency output value of the first target detection node, the second target detection node, up to the M-th target detection node;

[0032] The first target detection channel is activated to perform target recognition on the first image acquisition result to generate the first vehicle recognition result.

[0033] Furthermore, this step also includes:

[0034] Obtaining basic camera information, wherein the basic camera information includes camera model information and camera deployment distance, where the camera deployment distance refers to the closest distance between the camera and the first driving area;

[0035] Loading road images based on the camera model information and the camera deployment distance to obtain the first training image set;

[0036] Vehicle identification is performed on the first training image set to generate the first vehicle identification image set.

[0037] Furthermore, this step also includes:

[0038] Loading road images based on the camera model information and the camera deployment distance to obtain a first initial training image set and collect light intensity record values;

[0039] The first initial training image set whose recorded values ​​of the collected light intensity are greater than or equal to the light intensity threshold is screened and set as the first training image set.

[0040] In a preferred embodiment, camera-related information is first collected to obtain basic camera information, including camera model information and camera deployment distance. The camera model information refers to the model and specifications of the camera device, such as the manufacturer, model, and camera performance characteristics (e.g., resolution, lens type, and shooting capabilities). The camera deployment distance refers to the distance between the camera device and the vehicle's driving trajectory in the first driving area. Secondly, the camera model information and deployment distance are used to adjust camera settings to ensure that the captured images are adapted to specific environmental conditions. First, based on the camera model information, parameters such as exposure time, aperture size, and focus distance are adjusted to suit the characteristics of the specific camera model. Then, the configured camera is used to capture road images, including vehicles, road signs, and traffic lights, to obtain a first initial training image set. During image acquisition, light intensity is measured and recorded to obtain recorded light intensity values. The recorded light intensity values ​​are then analyzed and compared with a preset light intensity threshold. Images with recorded light intensity values ​​greater than or equal to the threshold are selected to form the first training image set, which helps ensure consistent lighting conditions for the images in the training dataset. At the same time, manually identify all vehicles in the first training image set, for example, by creating bounding boxes to clearly indicate the location of the vehicles in the images, thereby obtaining a first vehicle identification image set. Repeat the above process to obtain k sets of one-to-one correspondence between the first training image set and the first vehicle identification image set, forming the first training dataset.

[0041] Then, for a corresponding set of first training images and first vehicle identification image sets, randomly select an image from the first training image set and a corresponding vehicle identification image from the first vehicle identification image set, and add this pair of images to the first reconstructed training dataset. Perform k random extractions with replacement on each of the k corresponding sets of first training images and first vehicle identification image sets, each independent random extraction, to obtain the first reconstructed training dataset, which contains the k extracted image pairs.

[0042] Next, load the first reconstructed training dataset, which will be used to train the convolutional neural network; select an appropriate convolutional neural network architecture, such as ResNet, YOLO, SSD, etc.; use random weight initialization to initialize the selected convolutional neural network as the starting point for training. Subsequently, use the first reconstructed training dataset to perform multiple rounds of training. In each round, the image data will be fed into the network for forward propagation, and then the loss function is calculated to measure the difference between the output of the model and the actual identification. Furthermore, the weights of the neural network are adjusted according to the gradient of the loss function through the backpropagation algorithm to minimize the loss. The training process is repeated multiple times until the network converges and the first target detection node is generated. In the above manner, the second reconstructed training dataset is obtained until the Mth reconstructed training dataset is generated. The first target detection node is generated by the first reconstructed training dataset, and this is repeated M times. The second target detection node is generated in sequence from the second reconstructed training dataset to the Mth reconstructed training dataset.

[0043] Next, prepare the first, second, and Mth target detection nodes, each capable of detecting target vehicles in the image. Treating these nodes as parallel means they process the image simultaneously and generate their own detection results. The output of each node is a set of possible target vehicle locations and related information, serving as the first target detection channel. Next, for each node's output, a distance threshold is used to cluster the target vehicles that are close to each other in the detection results of different nodes. For each cluster, the number of nodes contained in it is calculated. The cluster with the largest number of nodes is considered the most frequent detection result and serves as the highest-frequency output value for the first, second, and Mth target detection nodes. This highest-frequency output value is then used as the output of the first target detection channel, providing reliable detection results. By combining the outputs of multiple target detection nodes into a single target detection channel and selecting the most frequently occurring target vehicle locations as the final result, this helps reduce false detection rates and improve the stability of target detection.

[0044] The generated first object detection channel is then activated, and a first image acquisition result is obtained from the camera. This first image acquisition result is then input into the first object detection channel to detect the target vehicle in the first image acquisition result. By applying the first object detection channel, the vehicle position in the first image acquisition result is detected, represented by a bounding box or other identifier, and features of the target vehicle, such as shape, color, and size, are extracted, thereby obtaining a first vehicle recognition result.

[0045] When the real-time light intensity is less than the light intensity threshold, the tracking radar is activated to perform image acquisition, a second image acquisition result is obtained, a second target detection channel is activated to perform target recognition, and a second vehicle recognition result is generated;

[0046] Furthermore, the embodiment of the present application also includes:

[0047] Acquire a second training data set, wherein the second training data set includes L sets of one-to-one corresponding second training image sets and second vehicle identification image sets;

[0048] Training the second object detection channel based on L sets of one-to-one corresponding second training image sets and second vehicle identification image sets;

[0049] The training process of the second target detection channel is the same as the training process of the first target detection channel.

[0050] In a preferred embodiment, a second training dataset is first collected using radar equipment. The second training dataset includes L sets of one-to-one corresponding second training images and a second vehicle identification image set, which are used to train the second object detection channel. The images in the second training image set include images below a light intensity threshold and driving scenes to ensure accurate operation of the second object detection channel. The images in the second vehicle identification image set correspond one-to-one with the images in the second training image set and include relevant information about vehicles appearing in the second training image set, such as vehicle locations, bounding boxes, and other landmarks.

[0051] The second target detection channel is then trained using the acquired second training dataset. The training process for the second target detection channel is identical to that for the first channel. L sets of one-to-one correspondences between the first training image set and the first vehicle identification image set are randomly sampled J times with replacement, repeated M times to generate M reconstructed training datasets. The convolutional neural network is then trained to generate M target detection nodes. These M target detection nodes are then merged to generate the second target detection channel. Subsequently, when the real-time light intensity falls below the light intensity threshold, the tracking radar is activated to acquire images, obtaining the second image acquisition result. The second target detection channel is then activated for target recognition, generating the second vehicle recognition result.

[0052] The multiple vehicle identification results of the multiple vehicle tracking components are spliced ​​together to generate a vehicle tracking result, wherein the first vehicle identification result and / or the second vehicle identification result belong to the multiple vehicle identification results.

[0053] Further, such as Figure 2 As shown, the embodiment of the present application also includes:

[0054] Clustering the plurality of vehicle recognition results according to the first target vehicle to obtain a plurality of first target vehicle tracking positions;

[0055] Sorting the plurality of first target vehicle tracking positions in time sequence to generate a first target vehicle tracking position sequence;

[0056] Based on the first target vehicle tracking position sequence and in combination with the traffic road topology, a first target vehicle tracking path is obtained and added to the vehicle tracking result.

[0057] In a preferred embodiment, multiple vehicle identification results are obtained using multiple vehicle tracking components. One of the vehicle identification results is selected as a first target vehicle and used as a basis for clustering the multiple vehicle identification results. Subsequently, a clustering algorithm, such as K-means clustering, is used to compare the multiple vehicle identification results and group them into different groups based on the identification result for the first target vehicle, thereby obtaining multiple first target vehicle tracking positions for the first target vehicle.

[0058] Each of the multiple tracked locations of the first target vehicle includes a timestamp and the vehicle's spatial coordinates. These locations are sorted by timestamp to ensure they are arranged in chronological order according to the vehicle's movement. This sorting results in a time-ordered position sequence, which serves as the first target vehicle tracking position sequence. This sequence contains the coordinates of the first target vehicle at different points in time and describes its trajectory, including its starting position, movement path, and final position.

[0059] Subsequently, a traffic road topology is retrieved, which describes the layout, connections, intersections, road signs, and lane information of the roads within the first driving area. The first target vehicle's tracking position sequence is matched with the traffic road topology to obtain the first target vehicle's complete motion path, which is used as the first target vehicle tracking path. This path includes information such as the first target vehicle's route, speed, and direction. Finally, the first target vehicle tracking path is added to the vehicle tracking results to achieve stable tracking of the target vehicle.

[0060] In summary, the vehicle tracking method integrating camera and radar provided in the embodiments of the present application has the following technical effects:

[0061] Multiple vehicle tracking components are deployed at preset intervals in the first driving area. Each of these components includes a tracking radar, a camera, and a light sensor, providing diversity and selectability for subsequent vehicle tracking to address target tracking needs under varying lighting and environmental conditions. The light sensor senses the real-time light intensity and determines whether it is greater than or equal to a light intensity threshold. Based on the real-time lighting conditions, the system determines which sensor to use for target recognition. When the real-time light intensity is greater than or equal to the light intensity threshold, the camera is activated for image acquisition, obtaining a first image acquisition result. The first target detection channel is activated for target recognition, generating a first vehicle recognition result. This ensures that the camera is used for target recognition under good lighting conditions to improve target recognition accuracy. When the real-time light intensity is less than the light intensity threshold, the tracking radar is activated for image acquisition, obtaining a second image acquisition result. The second target detection channel is activated for target recognition, generating a second vehicle recognition result. In poor lighting conditions, the system switches to the radar to obtain stable target tracking results to address environments with large lighting variations. Multiple vehicle identification results from multiple vehicle tracking components are spliced ​​together to generate a vehicle tracking result, wherein the first vehicle identification result and / or the second vehicle identification result belong to the multiple vehicle identification results. The identification results obtained by different sensors are integrated to provide more comprehensive and reliable vehicle tracking information, thereby improving tracking performance.

[0062] Example 2

[0063] Based on the same inventive concept as the vehicle tracking method integrating camera and radar in the aforementioned embodiment, Figure 3 As shown, an embodiment of the present application provides a vehicle tracking system integrating a camera and a radar, the system comprising:

[0064] A tracking component deployment unit 11 is configured to deploy a plurality of vehicle tracking components at preset intervals in a first driving area, wherein any one of the plurality of vehicle tracking components includes a tracking radar, a camera, and a light sensor;

[0065] The light intensity judging unit 12 is configured to sense the real-time light intensity through the light sensor and judge whether the real-time light intensity is greater than or equal to a light intensity threshold;

[0066] The first vehicle recognition unit 13 is configured to start the camera to perform image acquisition, obtain a first image acquisition result, activate a first target detection channel to perform target recognition, and generate a first vehicle recognition result when the real-time light intensity is greater than or equal to the light intensity threshold;

[0067] The second vehicle identification unit 14 is configured to, when the real-time light intensity is less than the light intensity threshold, start the tracking radar to perform image acquisition, obtain a second image acquisition result, activate the second target detection channel to perform target recognition, and generate a second vehicle recognition result;

[0068] The vehicle tracking result unit 15 is configured to concatenate the multiple vehicle identification results of the multiple vehicle tracking components to generate a vehicle tracking result, wherein the first vehicle identification result and / or the second vehicle identification result belong to the multiple vehicle identification results.

[0069] Furthermore, the first vehicle identification unit 13 includes the following execution steps:

[0070] Collecting a first training data set, wherein the first training data set includes k sets of one-to-one corresponding first training image sets and first vehicle identification image sets;

[0071] Perform k random samplings with replacement on k groups of one-to-one corresponding first training image sets and first vehicle identification image sets to generate a first reconstructed training data set;

[0072] Retrieving the first reconstructed training data set, training a convolutional neural network, and generating a first target detection node;

[0073] Repeat M times to obtain the second target detection node until the Mth target detection node;

[0074] Merging the first target detection node, the second target detection node, up to the M-th target detection node as parallel nodes to generate the first target detection channel, wherein the output value of the first target detection channel is the highest frequency output value of the first target detection node, the second target detection node, up to the M-th target detection node;

[0075] The first target detection channel is activated to perform target recognition on the first image acquisition result to generate the first vehicle recognition result.

[0076] Furthermore, the first vehicle identification unit 13 further includes the following execution steps:

[0077] Obtaining basic camera information, wherein the basic camera information includes camera model information and camera deployment distance, where the camera deployment distance refers to the closest distance between the camera and the first driving area;

[0078] Loading road images based on the camera model information and the camera deployment distance to obtain the first training image set;

[0079] Vehicle identification is performed on the first training image set to generate the first vehicle identification image set.

[0080] Furthermore, the first vehicle identification unit 13 further includes the following execution steps:

[0081] Loading road images based on the camera model information and the camera deployment distance to obtain a first initial training image set and collect light intensity record values;

[0082] The first initial training image set whose recorded values ​​of the collected light intensity are greater than or equal to the light intensity threshold is screened and set as the first training image set.

[0083] Furthermore, the second vehicle identification unit 14 includes the following execution steps:

[0084] Acquire a second training data set, wherein the second training data set includes L sets of one-to-one corresponding second training image sets and second vehicle identification image sets;

[0085] Training the second object detection channel based on L sets of one-to-one corresponding second training image sets and second vehicle identification image sets;

[0086] The training process of the second target detection channel is the same as the training process of the first target detection channel.

[0087] Furthermore, the vehicle tracking result unit 15 includes the following execution steps:

[0088] Clustering the plurality of vehicle recognition results according to the first target vehicle to obtain a plurality of first target vehicle tracking positions;

[0089] Sorting the plurality of first target vehicle tracking positions in time sequence to generate a first target vehicle tracking position sequence;

[0090] Based on the first target vehicle tracking position sequence and in combination with the traffic road topology, a first target vehicle tracking path is obtained and added to the vehicle tracking result.

[0091] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0092] Furthermore, the terms "first" or "second" as described above may not only represent an order relationship but may also represent a specific concept and / or refer to the selection of multiple elements individually or collectively. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, if such modifications and variations fall within the scope of this application and its equivalents, this application is intended to include such modifications and variations.

Claims

1. A vehicle tracking method integrating camera and radar, characterized in that: include: Deploying a plurality of vehicle tracking components at preset intervals in a first driving area, wherein any one of the plurality of vehicle tracking components includes a tracking radar, a camera, and a light sensor; The light sensor senses the real-time light intensity and determines whether it is greater than or equal to a light intensity threshold; When the real-time light intensity is greater than or equal to the light intensity threshold, starting the camera to perform image acquisition, obtaining a first image acquisition result, activating a first target detection channel to perform target recognition, and generating a first vehicle recognition result; When the real-time light intensity is less than the light intensity threshold, the tracking radar is activated to perform image acquisition, a second image acquisition result is obtained, a second target detection channel is activated to perform target recognition, and a second vehicle recognition result is generated; The multiple vehicle identification results of the multiple vehicle tracking components are spliced ​​together to generate a vehicle tracking result, wherein the first vehicle identification result and / or the second vehicle identification result belong to the multiple vehicle identification results.

2. The method according to claim 1, wherein When the real-time light intensity is greater than or equal to the light intensity threshold, the camera is started to perform image acquisition, a first image acquisition result is obtained, a first target detection channel is activated to perform target recognition, and a first vehicle recognition result is generated, including: Collecting a first training data set, wherein the first training data set includes k sets of one-to-one corresponding first training image sets and first vehicle identification image sets; Perform k random samplings with replacement on k groups of one-to-one corresponding first training image sets and first vehicle identification image sets to generate a first reconstructed training data set; Retrieving the first reconstructed training data set, training a convolutional neural network, and generating a first target detection node; Repeat M times to obtain the second target detection node until the Mth target detection node; Merging the first target detection node, the second target detection node, up to the M-th target detection node as parallel nodes to generate the first target detection channel, wherein the output value of the first target detection channel is the highest frequency output value of the first target detection node, the second target detection node, up to the M-th target detection node; The first target detection channel is activated to perform target recognition on the first image acquisition result to generate the first vehicle recognition result.

3. The method according to claim 2, wherein Acquiring a first training data set also includes: Obtaining basic camera information, wherein the basic camera information includes camera model information and camera deployment distance, where the camera deployment distance refers to the closest distance between the camera and the first driving area; Loading road images based on the camera model information and the camera deployment distance to obtain the first training image set; Vehicle identification is performed on the first training image set to generate the first vehicle identification image set.

4. The method according to claim 3, wherein Loading road images based on the camera model information and the camera deployment distance to obtain the first training image set includes: Loading road images based on the camera model information and the camera deployment distance to obtain a first initial training image set and collect light intensity record values; The first initial training image set whose recorded values ​​of the collected light intensity are greater than or equal to the light intensity threshold is screened and set as the first training image set.

5. The method according to claim 1, wherein When the real-time light intensity is less than the light intensity threshold, the tracking radar is started to perform image acquisition, a second image acquisition result is obtained, a second target detection channel is activated to perform target recognition, and a second vehicle recognition result is generated, including: Acquire a second training data set, wherein the second training data set includes L sets of one-to-one corresponding second training image sets and second vehicle identification image sets; Training the second object detection channel based on L sets of one-to-one corresponding second training image sets and second vehicle identification image sets; The training process of the second target detection channel is the same as the training process of the first target detection channel.

6. The method according to claim 1, wherein The multiple vehicle identification results of the multiple vehicle tracking components are combined to generate a vehicle tracking result, including: Clustering the plurality of vehicle recognition results according to the first target vehicle to obtain a plurality of first target vehicle tracking positions; Sorting the plurality of first target vehicle tracking positions in time sequence to generate a first target vehicle tracking position sequence; Based on the first target vehicle tracking position sequence and in combination with the traffic road topology, a first target vehicle tracking path is obtained and added to the vehicle tracking result.

7. The vehicle tracking system integrating camera and radar is characterized by: A vehicle tracking method for implementing the camera-radar fusion method according to any one of claims 1 to 6, the system comprising: A tracking component deployment unit, the tracking component deployment unit being configured to deploy a plurality of vehicle tracking components at preset intervals in a first driving area, wherein any one of the plurality of vehicle tracking components comprises a tracking radar, a camera, and a light sensor; a light intensity determination unit, configured to sense the real-time light intensity through the light sensor and determine whether the real-time light intensity is greater than or equal to a light intensity threshold; a first vehicle identification unit, configured to, when the real-time light intensity is greater than or equal to the light intensity threshold, start the camera to perform image acquisition, obtain a first image acquisition result, activate a first target detection channel to perform target recognition, and generate a first vehicle recognition result; a second vehicle identification unit, configured to, when the real-time light intensity is less than the light intensity threshold, start the tracking radar to perform image acquisition, obtain a second image acquisition result, activate a second target detection channel to perform target recognition, and generate a second vehicle recognition result; A vehicle tracking result unit is used to splice multiple vehicle identification results of the multiple vehicle tracking components to generate a vehicle tracking result, wherein the first vehicle identification result and / or the second vehicle identification result belong to the multiple vehicle identification results.

Images