Local Feature Of 3D Range Image For Ground Target Recognition

3D Imaging Lidar is widely used in military and civilian fields such as precision guidance,ground remote sensing,navigation and anti-terrorism security inspection because of its advantages such as long detection distance,accurate positioning information,multi-dimensional data dimension and strong anti-jamming ability.Although the ground rigid target recognition using local features can effectively solve the problem of target recognition of occlusion,it is easy to be affected by noise and varying data resolution when acquiring ground scene information by3 D Imaging Lidar,which leads to the decline in ground target recognition accuracy.Therefore,this paper focuses on the technology of ground target recognition based on local features.The main contents are as follows:（1）The key technologies in 3D ground target recognition such as model library establishment,keypoints extraction,feature description,feature matching and hypothesis generation and verification are deeply explored,and the influence of common methods of each step on the recognition performance is analyzed.The lidar is used to collect the actual data and make the ground targets database,which makes up for the deficiency in the ground rigid target of the model base.The existing construction methods of typical local feature descriptors are analyzed in depth,and the effects of construction methods such as local reference frame,geometric information coding and spatial partition on the performance of descriptors are clarified.it provides a theoretical support for the subsequent construction of local feature descriptors and the improvement of recognition accuracy.（2）In order to improve the ability of the local feature descriptor to describe the target feature in the case of noise interference,occlusion and the change of data resolution,this paper fully encodes the local surface space and geometric information in the local reference frame,combined with the information between keypoints,a local feature descriptor based on deviation angle statistics（Deviation angle statistics of keypoints from local points,adjacent keypoints,DASKL）is constructed.By pre-calculating the multi-scale local reference axis and selecting the appropriate local reference axis according to the scale strategy in the matching stage,the robustness to noise and different data resolution is improved,and the deviation angles for the local reference axis and the normal are fully encoded in the subdivision space to improve the description of the local surface.The deviation angle between the keypoint and the local reference frame of the three nearest keypoints is calculated to enhance the ability of the descriptor to distinguish similar local surfaces.Feature matching experiments is carried out in B3 R,UWAOR public database and ground target database,and compared with FPFH,SHOT and ROPS descriptors.The results show that the local feature descriptors proposed in this paper are more descriptive and timely,and are more robust to noise interference,occlusion and point cloud resolution changes.（3）In order to improve the accuracy of ground target recognition in the case of noise interference,occlusion and the change of data resolution,the cloth simulation filter is used to segment the ground scene in the recognition algorithm,combined with the descriptor of this paper.The methods of subsequent feature matching and hypothesis generation and verification are designed.In three databases,compared with the recognition algorithms based on FPFH,SHOT and ROPS descriptors,the experimental results show that the proposed method has better recognition results for noise interference,occlusion and data resolution changes.（4）In order to realize the application of ground target recognition technology in real scene,a ground targets recognition software prototype system is designed and developed,which integrates the relevant algorithms of this recognition method.An interactive interface that can change the important parameters of the algorithm and the interface to visualize the results of the algorithm are designed.In the system,the real ground scene and low-resolution target data are collected,and the recognition algorithm based on DASKL descriptor is used for real-time target recognition.Under the same conditions,it is compared with the recognition algorithm based on FPFH,SHOT and ROPS descriptors.The experimental results show the accuracy of this method in the practical application scene and the versatility for low-resolution targets.In order to solve the problem that the recognition accuracy of the 3D range images of ground scene decreases due to noise interference and the change of data resolution,a ground target recognition method based on local feature descriptor based on deviation angle statistics is proposed in this paper.Based on this method,a ground targets recognition software prototype system is developed.Through related experiments,the robustness and applicability of the proposed method is verified,and the practical engineering application of the ground targets recognition system is promoted.