CN-122017879-A - Mobile target precision tracking method and system with cooperative radar detection and vision locking

CN122017879ACN 122017879 ACN122017879 ACN 122017879ACN-122017879-A

Abstract

The invention discloses a radar detection and vision locking cooperative moving target precision tracking method, and belongs to the technical field of intelligent unmanned platforms. The system comprises a laser radar, a visual identification camera, a double-master control heterogeneous computing unit, an Ackerman chassis and a triaxial holder. The laser radar scans and positions a moving object, sends the coordinate position of a target to the three-dimensional cloud platform to accurately determine the position of the target, and then a visual end adopts an adaptive algorithm based on HSV color space and area filtering in a received coordinate area to realize robust target identification under complex illumination conditions such as strong light, backlight and the like. The chassis adopts AMCL+DWA global-local double-layer path planning, and the pan-tilt-zoom double-closed-loop coordination is realized based on image error PD closed-loop control. The system can control tracking errors within 5cm in a dynamic environment of more than or equal to 3m/s, and supports single/multiple car modes, illumination drafts and 30% shielding scenes. Compared with a single sensor scheme, the positioning accuracy is improved by 80%, the delay is reduced by 35%, and the method can be widely applied to high-speed maneuvering scenes such as security inspection, object tracking and the like.

Inventors

CUI YANJUN
ZHOU KOUHUA
DING ZHENPING
FEI YUE
LI JIAYIN
HE KAIDI

Assignees

南京理工大学紫金学院

Dates

Publication Date: 20260512
Application Date: 20250928

Claims (10)

1. The method for tracking the moving target with precision by cooperating radar detection and visual locking is characterized by comprising the following specific steps of: s1, data acquisition and synchronization; Rigidly fixing a laser radar and a camera in the same coordinate system, collecting point cloud data and high-frame-rate image data of a target in real time, and completing millisecond synchronization through a hardware timestamp; s2, data preprocessing and feature extraction; Transmitting the acquired data to a double-master-control heterogeneous computing unit, filtering, clustering and denoising point clouds, performing distortion correction, HSV space conversion and edge enhancement on images, and respectively extracting the 3D centroid, the speed vector and the 2D contour characteristics of the target; s3, radar-vision fusion and track prediction; Performing space-time alignment and track fusion on the coarse-granularity position of the radar and the vision fine-trimming outline by using Federal Kalman filtering, outputting a high-frequency centimeter-level target pose, and predicting the position at the next moment based on a historical track; S4, double closed loop tracking and striking decision; According to the fusion pose, the chassis performs closed-loop motion control of global-local double-layer path planning based on self-adaptive Monte Carlo positioning AMCL and dynamic window method DWA, and the cradle head performs PD closed-loop pointing control based on image errors, so that double closed-loop collaborative tracking of the vehicle body and the cradle head is realized; s5, recording and evaluating progress in real time; Recording each position, speed, tracking error and system decision log of the target in a database, and correlating the record with a time stamp to form a training and actual combat history record; S6, fatigue and fault monitoring; And according to the continuous operation time length, the temperature rise of the motor, the residual capacity of the battery and the communication heartbeat state, the fatigue grade of the system is evaluated in real time, and the function decline caused by overload is prevented.
2. The method for tracking the moving target with the cooperative detection and visual locking of the radar according to claim 1 is characterized in that the laser radar in S1 is a 16-line or more three-dimensional laser radar, the scanning frequency is more than or equal to 20Hz, and the cradle head reaction time is less than 200ms.
3. The method for accurately tracking a moving target by cooperation of radar detection and visual locking according to claim 1, wherein the data in S1 comprises 3D point cloud coordinates, reflection intensity, 2D pixel coordinates, time stamp and sensor internal and external parameter matrix of the target.
4. The method of claim 1, wherein the step of preprocessing the data in S2 includes point cloud voxel filtering, statistical outlier removal, image distortion correction, illumination normalization and time synchronization.
5. The method for precisely tracking the moving target by the cooperation of radar detection and visual locking according to claim 1, wherein the track is classified in the target motion mode in the step S3, and the track comprises linear motion, curve maneuver and emergency stop.
6. The method for accurately tracking a moving object in coordination with radar detection and visual lock according to claim 1, wherein the classification of the scene in S3 with the tracking difficulty level includes simple, medium and difficult.
7. The method for accurately tracking a moving object by cooperation of radar detection and visual lock according to claim 1, wherein the scene is classified by the number of tracking objects in S3, and the method comprises a single object and a plurality of objects.
8. A method for tracking movement target accuracy in cooperation with visual lock according to claims 1-7, wherein S4 further comprises the steps of: S41, generating a personalized strategy; comparing the current tracking performance with the historical optimal record according to the task target set by the operator; Marking the situation that the tracking error is larger or the number of times of unlocking is larger, and generating specific adjustment suggestions of the chassis speed, the cradle head gain and the prediction time domain; S42, continuously optimizing the model; And the fusion filtering and motion control parameters are continuously optimized by adopting an online incremental learning mode, so that the adaptability of the system to new scenes is improved.
9. The method for tracking the accuracy of a moving object in cooperation with the detection and the visual lock of claim 1, wherein S5 further comprises the steps of: s51, recording detailed data of each tracking task in a database, wherein the detailed data comprise target pose, tracking error, decision instruction and ambient light; S52, calculating a technical progress index of the system by analyzing task history records, wherein the technical progress index comprises the reduction rate of average positioning errors and the lifting amplitude of the maximum tracking speed, and presenting the technical progress to an operator in a chart and graph form by using a data visualization tool; The operator may set specific tracking goals and performance milestones and track the progress of the system in achieving these goals, providing immediate feedback and advice.
10. A high precision tracking system for radar detection and visual locking of moving objects, using the method of any one of claims 1-9 to track objects, comprising: The data acquisition module is rigidly arranged above the vehicle body, comprises a laser radar, a three-dimensional servo cradle head, a camera and a laser module, and is used for acquiring 3D point cloud and 2D image data of a target in real time and guaranteeing time synchronization of the data; the fusion decision module is positioned in the raspberry group and is responsible for receiving, storing and processing the acquired data, and comprises a radar-vision fusion sub-module, a personalized strategy generation sub-module, a task recording and history tracking sub-module and a fatigue and fault monitoring sub-module; the radar-vision fusion sub-module is used for generating a centimeter-level target pose and predicting a future track based on the point cloud and the image characteristics; The personalized strategy generation sub-module outputs parameter adjustment suggestions of the chassis and the cradle head according to the task target and the current performance; The task recording and history tracking sub-module is used for maintaining the history data of all tasks, including target tracks, tracking errors and decision logs, and allowing operators to check and analyze at any time; The fatigue and fault monitoring sub-module is used for calculating the fatigue index of the system in real time and sending out an alarm and a degradation strategy suggestion when the system is abnormal; The execution module comprises an Ackerman chassis, a triaxial holder and an optional laser/electromagnetic/net gun striking unit, and is used for completing movement of a vehicle body and target pointing according to the instruction of the fusion decision module; the display and interaction module comprises a computer webpage part and is used for providing real-time tracking pictures, error curves, fatigue early warning, social sharing and collaborative scheduling functions for operators.

Description

Mobile target precision tracking method and system with cooperative radar detection and vision locking Technical Field The invention relates to the technical field of intelligent mobile platforms, in particular to an unmanned vehicle system which integrates laser radar detection and visual locking and can realize centimeter-level high-precision tracking and autonomous striking on a high-speed moving target in a dynamic environment. Background The existing unmanned vehicle system on the market is mainly divided into three types according to different perception and decision-making architectures, namely vision leading type relies on a camera and a deep learning algorithm, the cost is low, but the positioning error is often more than 10cm under strong light, backlight or shielding environments, radar leading type adopts laser radar point cloud clustering, illumination interference is strong, but texture information is lacking, the target contour is difficult to identify, loose coupling fusion type simply splices the two results at a target or track layer, the time synchronization precision is only up to hundred milliseconds, and the error is rapidly amplified during high-speed operation. Therefore, under the working conditions of high-speed and complex illumination and shielding, the existing architecture still cannot meet the tracking requirements of centimeter-level precision and millisecond-level delay at the same time. When the two are independently operated, three requirements of high-speed movement, high-precision locking and strong robustness are difficult to meet at the same time. In recent years, although radar-vision fusion research exists, most of the radar-vision fusion research stays in static or low-speed scenes, and a system-level solution is still lacking for high-speed maneuvering scenes such as countermeasure, security patrol, logistics transportation and the like. Therefore, a high-precision tracking vehicle which is completely innovative in terms of hardware architecture, fusion algorithm, closed-loop control and the like is needed. Disclosure of Invention The invention provides a method and a system for precisely tracking a moving target by cooperative radar detection and visual locking, which aim to realize centimeter-level (less than or equal to 10 cm) real-time positioning and continuous locking of a high-speed moving target (more than or equal to 3 m/s), evaluate the fatigue state of a system in real time, feed back the fatigue state to technicians, prevent equipment from overheating, improve the working efficiency through the cooperative function of social sharing and multiple vehicles, and finally promote the intelligent and personalized development of multiple scenes such as intelligent tracking, target hitting and the like. A first aspect of the present invention provides a method for tracking a moving target with precision in cooperation with radar detection and visual locking, comprising the steps of: the method for tracking the moving target with precision by cooperating radar detection and visual locking is characterized by comprising the following specific steps of: s1, data acquisition and synchronization; Rigidly fixing a laser radar and a camera in the same coordinate system, collecting point cloud data and high-frame-rate image data of a target in real time, and completing millisecond synchronization through a hardware timestamp; s2, data preprocessing and feature extraction; the system adopts a double-master control heterogeneous computing unit and consists of a raspberry group 5 (responsible for image recognition, target tracking and hitting decision) and a raspberry group 4 (responsible for radar data processing and chassis control). Transmitting the acquired data to a raspberry pie 5, filtering, clustering and denoising point cloud, performing distortion correction, HSV space conversion and edge enhancement on an image, and respectively extracting a 3D centroid, a speed vector and 2D contour characteristics of a target; s3, radar-vision fusion and track prediction; Performing space-time alignment and track fusion on the coarse-granularity position of the radar and the vision fine-trimming outline by using Federal Kalman filtering, outputting a high-frequency centimeter-level target pose, and predicting the position at the next moment based on a historical track; S4, double closed loop tracking and striking decision; According to the fusion pose, the chassis executes closed-loop motion control of AMCL (self-adaptive Monte Carlo positioning) +DWA (dynamic window method) global-local double-layer path planning, the cradle head executes PD closed-loop pointing control based on image errors, and double closed-loop collaborative tracking of the car body and the cradle head is realized; s5, recording and evaluating progress in real time; Recording each position, speed, tracking error and system decision log of the target in a database, and correlating the record with a time s