Research Of Multi-source Cooperative Moving Object Detection Technology

Moving object detection technology is an important research subject in the field of computer vision,it not only has an important role in research,but also plays an important role in promoting the research of the other fields in computer vision.The main task of the moving object detection is to extract the moving objects from the video sequences,which is the foundation and key of the subsequent processing of target recognition,tracking and behavior analysis in video sequences.According to whether the camera is moving,moving object detection can be divided into two categories: moving object detection in static background and moving object detection in dynamic background.In most practical applications,the camera is fixed,that is the static background,and its detection technology is relatively mature.In the dynamic background,the camera is moving,the monitoring scope is larger and the cost is lower.However,due to the movement of the camera,the background is constantly changing,at this time,the moving object detection algorithm in the static background is no longer applicable.The change of background increases the difficulty of moving object detection.In this thesis,the commonly used algorithms of moving object detection are intensively studied,and these in dynamic background are studied in details.Global motion compensation algorithm is usually used to compensate the motion of background caused by camera motion in the dynamic background,namely converting dynamic background into static background,and then the moving object detection algorithm in the static background is used to conduct the detection of foreground target.Therefore,the moving object detection algorithms in the static background are studied firstly in this thesis,especially for background subtraction.The global motion compensation method contains two parts: global motion vector estimation and background motion compensation.The core of the algorithm is to find the transform relation and then estimate the global motion vector according to the registration of the image sequences.The commonly used global motion vector estimation algorithm can be divided into two categories: one is based on image gray information,and the other is based on image features.The algorithm based on image gray information includes grayprojection method,block matching method,optical flow method and so on.The gray projection method and the block matching method can only be used to estimate the motion of the camera accurately and the computation of the optical flow method is large,so it is difficult to meet the real-time requirements.The estimation algorithm based on image feature requires a high degree of image feature selection.With the movement of the camera,some image features may disappear.Aiming at the above problems,to complete the registration of optical images,infrared images,optical and infrared images,the alignment degree is used as the matching criterion,and for particle swarm algorithm is easy to fall into the local extremum,the particle swarm optimization algorithm based on simulated annealing is used as the search strategy in this thesis.After converting the dynamic background to a static background,the moving object detection algorithm in the static background is used to detect moving objects.Due to the existence of “hole” in the moving object detection by the frame difference method,the background subtraction method is used to detect the moving target in this thesis.The core of the background subtraction method is to establish the background model,the extraction of moving objects is realized by using the method of visual background extraction with better performance in this thesis.Finally,to improve the integrity and accuracy of the detection results,firstly the moving object detection is respectively conducted on the optical and infrared image sequences and then the image fusion of decision level is conducted based on the detection results in this thesis.