Research On Key Technologies Of Moving Target Detection Under Camera Motion

The moving camera system is a real-time video processing system using a rapidly rotating and translating camera. Moving target detection is a very important part of digital image processing and in recent years, it has been a research hotspot. Moving target detection detects moving target from moving video sequences for further processing such as target searching, tracking and distinguishing. Moving target detection is widely applied in the field of security monitoring, intelligent robot vision, military aviation and so on. With the development of science and technology, the application of the Internet of things, moving target detection technology can also be used in handheld devices, so moving camera based moving target detection is an urgent problem to be solved. Because handheld terminal devices such as vehicle, often need to move in deep scene where strong parallax exists. Stationary objects closer to the camera would be judged as moving target so that false alarm appears near strong parallax. The real moving target cannot be detected correctly when false alarm appears, In that case, moving target detection rate decreases.In this paper, the moving target detection technology based on moving camera is used as the research object. Firstly, the research significance and application fields of the moving target detection technology, and the research status of domestic and foreign scholars are introduced. Secondly, basic principles of moving target detection technology are summarized. The derivation and estimation of machine vision coordinate system and pinhole camera model are analyzed and the development of moving target detection technology and the existing technology are emphatically analyzed. Thirdly, the disadvantages of the high false alarm rate in the presence of strong parallax are analyzed. Based on the "Plane + Parallax" framework, multi-planar parallax constraint is deduced. This paper also deduces moving target detection method based on multi-planar parallax constraint and verified the performance of the algorithm in strong parallax environment qualitatively and quantitatively with the actual video sequences. Experimental result shows that moving target detection method based on multi-planar parallax constraint has a good performance in the case of single target and multi target. The algorithm has certain robustness as well. Fourthly, this paper analyses the effect of 3-dimension depth information during the moving target detection process. As depth information is used in depth constraint, parallax could be completely eliminated from origin. This paper also deduces moving target detection method based on depth constraint and verified the performance of the algorithm in strong parallax environment qualitatively and quantitatively with the actual video sequences. Experimental result shows that moving target detection method based on depth constraint has a good performance in the case of single target and multi target.