A Study On Object Tracking Method Based On Three-dimensional Laser Data

As one of key technologies for the automatic attack in the future battlefield,the laser active imaging guidance technology can obtain three-dimensional(3D)data containing the structural information obtained by missile-borne LIDAR,with which can hit the taget more accurately.There are little studies using the above data obtained by missile-borne LIDAR to recognize and track the target.Based on the above situations,studies are carried out and the main contents of this paper are as follows:(1)Considering the moving LIDAR and its actual way of obtaining data on the missile platform,a dataset containing laser 3D data for object tracking is built and verified.Experiment shows the dataset is highly simulated.(2)Since that the non-ground points contains more valuable information,the cloth simulated filtering algorithm and clustering segmentation algorithm based on Euclidean distance are used to extract the non-ground points and clusters(for data segmentation)from the original data.And summed volume region(SVR)selection is proposed to exclude the cluster which is different from the interested target.Experimental results show that the above preprocessing methods can effectively reduce the following workload and improve the real-time performance of the algorithm.(3)Because that the performance of current feature descriptors decreases when the missile-object distance changes continuously,a joint global and local feature(JGLF)descriptor is proposed.And comparisons between the proposed descriptor and five other frequently-used descriptors are carried out in many aspects.The proposed JGLF descriptor achieves the balance between recognition accuracy and real-time performance in the experiments.(4)To track the object steadily when missile-object distance changes,the algorithm based on matching and recognizing and location and space structures are proposed respectively.With the proposed JGLF descriptor,frames of object recognition method based on fature matching and particle filter algorithm are used to build object tracking algorithms.Experiments show that the former has a high tracking accuracy but a relative bad real-time performance.The latter is not as accurate as the former but it performs faster.Therefore,combined with their respective advantages,both algorithms can be applied to the laser active imaging guidance process.The main innovations of this paper are as follows:(1)Aimed at object tracking,the SVR selection is proposed as a laser 3D preprocessing step.The pre-selection is applied to the corresponding data because of the size difference of real objects.This method is aimed to reduce the workload of subquent feature extraction and matching.(2)The JGLF descriptor is proposed to overcome the instability generated from the missile-object distance change.Using JGLF descriptor to track the target can offer more accurate and stable results and improve the real-time performance.