Study Of Rigid Deformation Target Recognition Based On 3D Point Cloud

Three-dimensional information is the essential attribute of the target, which has the stable and anti-jamming features. The object detection and recognition have been widely used in all areas of military, industrial, civil and other fields based on three-dimensional information. Relative to the two-dimensional image based on object recognition technology, three-dimensional object recognition technology still mostly stay in theory and method validation phase, yet it has not been applied massively, so in practice will encounter many difficulties and problems, especially the deformation that will occur rigid three-dimensional object detection and recognition, there is no practical solution. Battlefield Tank is a typical high-value targets, which has important military value in its detection and recognition, but the fact that the tank turret can rotate 360 degrees, belonging to the typical of the rigid deformations goal, at present domestic and overseas have no idea to the method of tank target which can correctly identify the rear turret of arbitrary rotation angle. In order to solve this problem, in this article, put forward and realized a three-dimensional deformation target recognition method based on segmentation, which is divided into a first target tank hull and turret two moving parts, and then apply the recognition method based on three-dimensional feature recognition components respectively, Finally, to fuse the parts of identification result, the ultimate goal which is to accurately identify the target model. The main work includes the following aspects:(1) In the aspects of three-dimensional point cloud data acquisition and processing, based on laser radar measurement model, having conducted the three-dimensional imaging simulation, you can get tanks at different rotation angles, three-dimensional point in different locations cloud data, having researched and implemented the cloud data processing method of point cloud filtering, ground point removal and target segmentation.(2) In terms of three-dimensional object segmentation, having implemented and improved the traditional growth-based and regional-based three-dimensional object segmentation algorithm minimum spanning tree, the experimental results show that these methods of error segmentation tanks will occur. According to the geometric characteristics of the tank, this article puts forward segmentation method based on Hough transform, tanks with different angle of rotation can be segmented well in this way.(3) In the three-dimensional object recognition, having adopted recognition methods based on three-dimensional feature, identifying the turret and body respectively, according to the similarity measure to identify the type of scene and scene correct template matching rate, the higher the recognition accuracy rate, and further can be calculated the relative angle of rotation of the turret and the body of experimental results show that the angle error is small.