Study Of Ground Target Recognition Based On Dual Wedge Scanning Laser Radar

Laser radar sensor has high ranging precision, it can obtain the three-dimensional reliable contours information of objects, which is an effective method to identify vehicle in complex terrain background. Under this background, the 3 D automatic target recognition technology based on laser radar was studied in this paper. The target recognition method based on spin-images was studied and the efficiency was improved. According to application requirements, a fast laser radar target recognition method based on projection dimension reduction was proposed in this paper. The main contents of the study are as follows:In the three dimensional ground target data acquisition and processing: firstly, the double wedge scanning mechanism was modeled. Then realize the airborne laser radar forward-looking perspective visual simulation under complex background, and obtain the point cloud data of three-dimensional object and scene. The processing method of point cloud data was discussed. Data filtering and classification and the extraction of three-dimensional target point cloud data were realized.In 3D target recognition algorithm based on laser radar, The target recognition method based on spin-image feature was studied. To adapt to the actuality that distribution of airborne scanning laser radar, sampling and homogenization were applied on the point cloud data. Selecting the key points and calculate the spin-image, so that the feature set for recognition was simplified. Finally the real-time performance and applicability of the algorithm was effectively improved.To meet the real-time requirements for ATR, we propose a three-dimensional ATR algorithm based on projection dimension reduction. The core idea of the algorithm is the use of overall shape features of target. Apply the projection dimension reduction method into the three-dimensional point cloud data extracted to get the surface projection map. A top-down matching algorithm was proposed. The combination of contour similarity measure value and the similarity of size measurement act as the matching recognition criteria for target recognition. The real-time performance of this method was improved by projection dimension reduction. Ground target recognition experiments were made to analysis the engineering performance of the method proposed. The results show that the method has better classification and identification capability for vehicle targets, and it was robust against target gesture and self-occlusion. The influences of point cloud spatial resolution, laser radar system noise and target motion, the blocking effects of sparse trees on target recognition were further studied.