| The technology of target tracking and positioning of airborne electro-optical system is a hot research topic in the field of machine vision.Its basic connotation refers to that under the platform of airborne electro-optical system,we detect moving targets on the ground and then recognize moving pedestrian.Then,through stable tracking of targets,combining with airborne information,using passive location technology,we can realize the positioning of the pedestrian’s geographic coordinates(latitude and longitude).By connecting to WJ icon drawing system,we can ultimately realize the positioning of pedestrians on object map,and that can provide direct viewing analysis basis for battles.This technology has important application value in the border patrol,NBC rescue etc..This paper is based on that the CAPF searching pedestrian targets,we focus on moving object detection,pedestrian detection,pedestrian tracking and geographical coordinates passive positioning of this system and we launch the research in these aspects.The main work and research contents include that:1.We researched a regional change moving target detection method based on Bhattacharyya coefficient.The method uses gray histogram information to select image sequence registration of feature points,uses affine transformation parameters model to recognize fast registration.By sharing registration image into 32× 32 pieces,calculating gray kernel function of each piece,using the Bhattacharyya coefficient as the kernel function of small change comparison tools,we can realize the moving target detection.Experiments show that the method can reduce the computational complexity of image registration,and overcome the effects of factors such as light detecting effects.After processing,the target area will be divided into 64×128 images,and provides accurate input image for the identification of pedestrians.2.A fast pedestrian recognition method based on HOG+ELM learning machine is proposed.The method mainly uses the advantages that the input connection weights of the network and the bias of the implicit elements of ELM learning machine to optimize the iterative process and reduce the computation time.Experimental results show that if we select targets of same features and ensure the correct recognition rate.This can provide the initial target tracking position for pedestrian tracking.3.Aiming at the problem that susceptible to occlusion and background clutter and other issues of pedestrian tracking process,a study based on the context of assisted pedestrian tracking algorithm has been researched.The method makes use of the context relationship of all the target pixels and targets around pedestrians,and then the tracking problem can be seen as a confidence map of the location of the pedestrian,confidence map in response to the largest location is the target center.The experimental results show that the proposed algorithm has strong anti blocking ability and adaptive background clutter compared with CT and other four methods,so it can achieve the stable tracking of the human target in airborne platform.In order to solve the problems of large calculating in coordinate transformation of traditional target positioning,we propose a method base on pixel passive position calculation.Combined with airborne GPS information,we can get the precise geographical coordinates of the target pedestrians,and then connect to WJ chart plotter system,so that we can achieve the goal of precise positioning of pedestrians on the electronic map and satellite three-dimensional map.This technology can provide a reliable basis for the field of intelligence analysis and emergency operations. |
