Research On Target Feature Extraction And Recognization Based On Imaging Ladar

In order to meet the demand of investigation, surveillance and precision-guide in the future battlefield environment, information processing of imaging laser radar has led an important research direction at home and abroad. Target recognition based on the data acquired by imaging laser radar is not only a key issue in this area, but also a challenging puzzle. This paper focuses on the target recognition problem of imaging laser radar, researchers here have launched the researches of imaging laser radar simulation, target orientation estimation, local feature descriptors and target recognition. The main work and results are as follows:1. To improve research efficiency and reduce test costs, this study acquires imaging laser radar echo data by simulation means. Firstly, the basic theory of laser radar simulation has been carefully studied, then comes the description of the laser beam emission, transmission, target reaction and the receiving process in mathematical models, thus to generate relatively realistic laser radar image of target. In the process of experiments, we found a problem of the simulation efficiency of scanning laser radar, and put forward a fast algorithm to solve this problem. We build up a mathematical model to describe the moving target and the scanning laser beam, then transformed the light-target intersection problem into an in-equation solving problem. Through the processing of equations in a piecewise linear way, find out possible solutions. Due to this method, we greatly streamlined the light-target intersection operations, and improved the efficiency of the simulation algorithm. Experimental results show that the proposed algorithm guarantees the fidelity of this simulation, and in the same time greatly improve the simulation efficiency.2. To solve the warhead orientation estimation problem, we proposed a orientation estimation algorithm for warhead. This article took the warhead as a cone, and deduced a mathematical model describing this cone. We utilizing the optimization algorithm to estimate the shape, orientation and location of the point cloud obtained by the laser radar, which is also the parameters of the warhead target. Experiment results show that the algorithm does not require a large number of points within a point cloud, and can achieve a high level of estimation performance at a high ranging accuracy.3. We studied the factors that affect the performance of traditional local feature descriptors, and pointed out that the descriptors' sensitivity to local reference frame(LRF) error is a crucial factor that lower the quality of its performance under various environment and in the situation of noise. In order to improve the descriptors' robustness to noise and varying mesh resolutions, we proposed a descriptor by utilizing spherical harmonic rotation invariants. The parameters required for extracting features has undergone a number of experiments to make it credible. We compared our descriptor against several traditional descriptors under different datasets, noises and mesh resolutions. The experimental results show that our descriptor outperformed other ones. A target recognition algorithm has also been put forward to test the recognizing ability of local feature descriptors, in which 5 different sorts of targets were used for experiments. And our descriptor achieved the highest recognition rate.