Target Detection And Tracking Technology Based On Gm-APD Lidar Range Image

With the laser lidar field development,research on imaging laser lidar has gradually entered people's field of vision.The Gm-APD array imaging lidar is highly valued by research institutions and experts due to its high sensitivity and long detection range.It can be used on missiles or aircraft to detect and track long-range targets.At present,three-dimensional target detection and recognition methods are mostly based on scanning lidar data,and their imaging characteristics are fundamentally different from APD.Therefore,related results cannot be directly used in this application.Based on the background above,this paper proposes and implements a set of algorithm flow based on Gm-APD lidar range image,including image restoration,detection,and tracking algorithms,and an outdoor demonstration and verification system has been built.Specific work includes:Aiming at the range image signal extraction under the condition of low signal-to-noise ratio,based on the analysis of the lidar imaging model,this paper proposes a weighted Gaussian smoothing linear regression signal restoration algorithm.First,Gaussian weighted smoothing is performed on the frequency distribution histogram of statistical imaging data.The noise in the strobing gate is gated,and the echo-response at the target is improved.Then use the linear regression method to segment the histogram,calculate the offset curve corresponding to the histogram,and restore the target's true position.Experiments prove that this method has high image restoration performance.Aiming at the target detection of the combined frame restoration range image,this paper proposes a target detection algorithm based on range image-point cloud fusion.On the range image,this paper uses the gradient operator and the threshold segmentation method to detect the target,based on data distribution characteristics of the target area and the background noise area,and frame the area for the point cloud detection.For the next step,the image is converted into a point cloud according to the lidar parameters and the range image,and the hierarchical merging DBSCAN algorithm is used on the point cloud to cluster the targets in the image quickly.Finally,the advantages and disadvantages of the two methods are combined to fuse the images,and an image containing only the target is obtained.To facilitate the next tracking step,this paper also designs a method based on the minimum bounding box to identify and select the final target and estimate the size,distance,and location of the target.The algorithm can fully and effectively detect the target in the range image through the target detection test on the data set.Aiming at the target tracking of the combined frame restoration range image,this paper aims at the shortcomings of the traditional KCF applied to the lidar range image.By making full use of the depth information of the range image,a KCF tracking algorithm based on the distance measurement is proposed.The algorithm can remove part of the influence of noise interference before tracking,and can adaptively adjust the size of the target bounding box;it can maintain the tracking model not polluted when the target is temporarily lost,and it can automatically re-enter the target detection stage when the target is completely lost,then initialize the detection and tracking process.Experiments prove that this method improves the stability of tracking,and can also prevent the impact caused by the short-term loss of the target.Finally,this article also designs the entire system and writes a prototype system that includes lidar hardware control,communication transmission,algorithm implementation,and other functions.