Single sensor multi-target association matching method, system and vehicle

Abstract

The present invention discloses a single-sensor multi-target association matching method, system and vehicle, including: step 1, obtaining target information of a driving road; step 2, preprocessing the obtained target information; step 3, performing time compensation and spatial synchronization on the detected target; step 4, generating a first detection SIFT feature map according to the position and speed of the detected target; step 5, performing time compensation and spatial synchronization on the tracked target; step 6, generating a first tracking SIFT feature map according to the position and speed of the tracked target; step 7, performing target association pairing based on the first detection SIFT feature map and the first tracking SIFT feature map; step 8, evaluating the optimal matching pair in combination with the tracking pairing information, and outputting the optimal matching pair as the detection target of the sensor. The present invention can realize multi-frame association tracking of a single sensor target to obtain target trajectory information, and can also realize target association between multi-source heterogeneous sensors to restore the real driving scene.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-sensor fusion, and in particular relates to a single-sensor and multi-sensor multi-target association matching method, system and vehicle. Background Art

[0002] With the vigorous development of the science and technology era, technological innovation has become the core driving force for all walks of life. The demand for automated products by mass consumers is increasing and becoming more and more widespread. However, for the automotive industry, autonomous driving is not only an important function to meet customer needs, but also an effective way to boost the brand. Therefore, autonomous driving in the automotive industry has become a new round of research boom under the background of the new era and new technology. With the current technical support of various industries, autonomous driving based on structured roads has become the main direction of this field, but there are still two major technical difficulties: 1. Multi-source sensor target association; 2. Multi-source sensor target attribute fusion. The complex real environment determines that a single sensor is insufficient to express reality. The multi-source heterogeneous sensor architecture has become the first choice for autonomous driving development mode, which plays a key role in the reconstruction of the vehicle driving environment. However, due to environmental and hardware limitations, different sensors will have false targets, loss, and jumps and errors in key information such as position and speed.

[0003] Therefore, it is necessary to develop a new single-sensor and multi-sensor multi-target association matching method, system and vehicle. Summary of the invention

[0004] The purpose of the present invention is to provide a single-sensor multi-target association matching method, system and vehicle, which can realize single-sensor target multi-frame association tracking to obtain target trajectory information.

[0005] Another object of the present invention is to provide a multi-sensor multi-target association matching method, system and vehicle, which can realize target association between multi-source heterogeneous sensors and restore the real driving scene.

[0006] In a first aspect, a single-sensor multi-target association matching method according to the present invention comprises the following steps:

[0007] Step 1: Obtain target information of the driving road, wherein the driving road target information includes a target tracking ID, a position, a speed, an acceleration, a target type, a motion trend, and a heading angle in a vehicle coordinate system;

[0008] Step 2: Preprocess the acquired target information to filter out invalid and obviously abnormal targets;

[0009] Step 3: Perform time compensation and spatial synchronization on the detected target;

[0010] Step 4: Generate a first detection SIFT feature map according to the position and speed of the detection target;

[0011] Step 5: Combine the target tracking characteristics of the sensor to perform time compensation and spatial synchronization on the tracked target;

[0012] Step 6: Generate a first tracking SIFT feature map according to the position and speed of the tracking target;

[0013] Step 7: performing target association pairing based on the first detection SIFT feature map and the first tracking SIFT feature map;

[0014] Step 8: Combine the target association pairing output in step 7 to evaluate the best matching pair, and output the best matching pair as the detection target of the sensor.

[0015] Optionally, in step 4, the position of the detected target is used as the main attribute key point, and the speed of the target is used as the pixel brightness to construct a first detection SIFT feature map.

[0016] Optionally, in step 6, the position of the tracked target is used as the main attribute key point, and the speed of the tracked target is used as the pixel brightness to construct a first tracking SIFT feature map.

[0017] In a second aspect, a multi-sensor multi-target association matching method according to the present invention is used to perform association matching on targets of multiple sensors, comprising the following steps:

[0018] Step 1: Single sensor multi-target association matching:

[0019] Using the single-sensor multi-target association matching method of the present invention, each single sensor outputs the best matching pair as the target detected by itself;

[0020] Step 2: Multi-sensor target association matching:

[0021] Construct a second detection sift feature map according to the target information detected by the multi-sensor;

[0022] Construct a second tracking sift feature map based on the target information tracked by the multi-sensor;

[0023] Performing target association pairing based on the second detection sift feature map and the second tracking sift feature map;

[0024] The optimal matching pair is evaluated by combining the fusion characteristics and is output as the detection target of the multi-sensor.

[0025] Optionally, a second detection SIFT feature map is constructed according to target information detected by multiple sensors, specifically:

[0026] According to the target information output by each single sensor, the eigenvalue of SIFT is calculated by weight distribution method to construct the second detection SIFT feature map.

[0027] Optionally, the second tracking SIFT feature map is constructed according to the target information tracked by the multi-sensor, specifically:

[0028] According to the target information tracked by each single sensor, the eigenvalue of SIFT is calculated by weight distribution method, and the second tracking SIFT feature map is constructed.

[0029] In a third aspect, a single-sensor multi-target association matching system described in the present invention comprises a memory and a controller, wherein the memory stores a computer-readable program, and when the computer-readable program is called by the controller, it can execute the steps of the single-sensor multi-target association matching method described in the present invention.

[0030] In a fourth aspect, a multi-sensor multi-target association matching system described in the present invention is characterized in that it includes a memory and a controller, the memory stores a computer-readable program, and the computer-readable program can execute the steps of the multi-sensor multi-target association matching method described in the present invention when called by the controller.

[0031] In a fifth aspect, a vehicle described in the present invention adopts the single-sensor multi-target association matching system described in the present invention.

[0032] In a sixth aspect, a vehicle described in the present invention adopts the multi-sensor multi-target association matching system described in the present invention.

[0033] The present invention has the following advantages:

[0034] (1) For the target association of a single sensor, it can form a target association list based on the tracking target association matching detection target, and effectively combine the context information of the target in the environment; it not only enhances the robustness and stability of the single sensor target, but also improves the credibility of the target's true attributes, enhances the efficiency and performance of the fusion, and reduces resource consumption;

[0035] (2) For multi-sensor target association, the fused tracking target can be effectively associated with the target detected by each sensor, which improves the authenticity of the fused target to the real scene, enhances the reliability of target fusion, and reduces the resource consumption of environmental cognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the single sensor multi-target association matching method in this embodiment;

[0037] Figure 2 4 is a flow chart of the multi-sensor multi-target association matching method in this embodiment. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with the accompanying drawings.

[0039] like Figure 1 As shown, in this embodiment, a single sensor multi-target association matching method includes the following steps:

[0040] Step 1: Obtain target information of the driving road, wherein the target information of the driving road includes target tracking ID, position, speed, acceleration, target type, motion trend and heading angle in the vehicle coordinate system. The single sensor can be any one of a camera, a millimeter wave radar and a laser radar;

[0041] Step 2: Preprocess the acquired target information, that is, specify preprocessing data cleaning rules according to different sensor characteristics and prior knowledge to filter out invalid and obviously abnormal targets.

[0042] Step 3: Combine the sensor measurement characteristics to perform time compensation and spatial synchronization on the detected target. Since there are algorithm and transmission delays from sensor perception to target output, it is necessary to compensate for the time difference and perform spatial synchronization on the delay time in combination with the delay characteristics of the perception model.

[0043] Step 4: Generate the first detection sift feature map according to the position and speed of the detection target, specifically:

[0044] The position of the detected target is taken as the main attribute key point, the speed of the target is taken as the pixel brightness, and the first detection SIFT feature map is constructed.

[0045] Step 5: Combine the target tracking characteristics of the sensor to perform time compensation and spatial synchronization on the tracked target.

[0046] Step 6: Generate the first sift feature map according to the position and speed of the tracking target (based on the historical detection target, calculated by the target tracking algorithm, the tracking target is the existing technology, which will not be repeated here), specifically:

[0047] The position of the tracked target is taken as the main attribute key point, and the speed of the tracked target is taken as the pixel brightness to construct the first tracking SIFT feature map.

[0048] Step 7: Based on the first detection SIFT feature map and the first tracking SIFT feature map, an image processing feature point matching algorithm is used to perform target association pairing and find matching pairs in the feature map.

[0049] Step 8: Combine the target association pairing output in step 7 to evaluate the best matching pair, and use the best matching pair as the sensor's detection target output. Use the feature data extracted during the tracking process to evaluate the confidence of the current matching pair, thereby enhancing the reliability of the single sensor output target.

[0050] In this embodiment, the confirmed pairing is removed from the pairing pool based on the historical matching pair information; and the optimality of the pairing is confirmed using the target position geometric relationship.

[0051] In this embodiment, the SIFT feature map includes the geometric position relationship between the feature target and the vehicle, the target speed, and the relationship between the targets.

[0052] like Figure 2 As shown, in this embodiment, a multi-sensor multi-target association matching method is used to associate and match targets of multiple sensors, including the following steps:

[0053] Step 1: Single sensor multi-target association matching:

[0054] Using the single-sensor multi-target association matching method described in this embodiment, each single sensor outputs the best matching pair as the target detected by itself.

[0055] Different sensors extract driving road target information in different ways. For example, cameras use deep learning algorithms to detect targets in lane images; millimeter wave and lidar use point cloud clustering algorithms to extract targets.

[0056] Step 2: Multi-sensor target association matching:

[0057] Construct a second detection sift feature map according to the target information detected by the multi-sensor;

[0058] Construct a second tracking sift feature map based on the target information tracked by the multi-sensor;

[0059] Performing target association pairing based on the second detection sift feature map and the second tracking sift feature map;

[0060] The optimal matching pair is evaluated by combining the fusion characteristics and is output as the detection target of the multi-sensor.

[0061] Optionally, a second detection SIFT feature map is constructed according to target information detected by multiple sensors, specifically:

[0062] According to the target information output by each single sensor, the eigenvalue of SIFT is calculated by weight distribution method to construct the second detection SIFT feature map.

[0063] Optionally, the second tracking SIFT feature map is constructed according to the target information tracked by the multi-sensor, specifically:

[0064] According to the target information tracked by each single sensor, the eigenvalue of SIFT is calculated by weight distribution method to construct the second tracking SIFT feature map.

[0065] In this embodiment, a single-sensor multi-target association matching system includes a memory and a controller, wherein the memory stores a computer-readable program, and when the computer-readable program is called by the controller, it can execute the steps of the single-sensor multi-target association matching method described in this embodiment.

[0066] In this embodiment, a multi-sensor multi-target association matching system is provided, characterized in that it includes a memory and a controller, wherein a computer-readable program is stored in the memory, and when the computer-readable program is called by the controller, the steps of the multi-sensor multi-target association matching method described in the present invention can be executed.

[0067] In this embodiment, a vehicle adopts the single-sensor multi-target association matching system as described in this embodiment.

[0068] In this embodiment, a vehicle adopts the multi-sensor multi-target association matching system as described in this embodiment.

Claims

1. A single sensor multi-target association matching method, characterized in that: The following steps are involved: Step 1: Obtain target information of the driving road, wherein the driving road target information includes a target tracking ID, a position, a speed, an acceleration, a target type, a motion trend, and a heading angle in a vehicle coordinate system; Step 2: Preprocess the acquired target information to filter out invalid and obviously abnormal targets; Step 3: Perform time compensation and spatial synchronization on the detected target; Step 4: Generate the first detection sift feature map according to the position and speed of the detection target, specifically: The position of the detected target is taken as the main attribute key point, the speed of the target is taken as the pixel brightness, and the first detection sift feature map is constructed; Step 5: Combining the target tracking characteristics of the sensor, time compensation and spatial synchronization are performed on the tracked target; the tracked target is a historically detected target; Step 6: Generate the first tracking sift feature map according to the position and speed of the tracking target, specifically: The position of the tracked target is taken as the main attribute key point, and the speed of the tracked target is taken as the pixel brightness to construct the first tracking sift feature map; Step 7: performing target association pairing based on the first detection SIFT feature map and the first tracking SIFT feature map; Step 8: Combine the target association pairing output in step 7 to evaluate the best matching pair, and output the best matching pair as the detection target of the sensor.

2. A single sensor multi-target association matching system, characterized by: It comprises a memory and a controller, wherein a computer-readable program is stored in the memory, and when the computer-readable program is called by the controller, the steps of the single-sensor multi-target association matching method as claimed in claim 1 can be executed.

3. A vehicle, characterized in that: A single sensor multi-target association matching system as described in claim 2 is adopted.

Images