Removed Object Detection Under Complex Environments

Removed Object detection is one of the important aspects of Intelligent Video Surveillance, which combines the advanced technologies of image processing, video analysis, artificial intelligence, pattern recognition and so on. It both can be applied to museums, exhibition halls and other public places and also be for key cultural sites, ordnance powerhouse and other privacy units.This thesis mainly studies the removed object detection based on static monocular camera, aims to detect one or more targets in the region of interest (ROI), once the target has been removed and the time of the target disappearance beyond the preset time, it immediately start-up a alarm. The main works are:Firstly, we use interactive way to get the ROI of the target to be detected. The setting of ROI is one to one, we can monitor one or more target at the same time. Under demands, the users can modify, add and remove a ROI real-time.Secondly, we uniting the advantages of Gaussian mixture background model and the time-mean background model, mainly based on the Gaus-sian mixture background model, within the time when there is light muta-tion,we use time-mean background model to detect the target, and real-time update the Gaussian mixture background model, when the model is stable, we use the Gaussian mixture background model again for foreground detection.Thirdly, based on the difference of scene complexity, we proposed two methods for distinguish the event of target occluded and removal:First, based on brightness information, we use the pixel's mean value, this method is simple, high efficiency and suitable for simple scene. Second, we use the seed filling algorithm based on color information, experiments show that this method is suitable for complexity scene and have better stability. Finally, a large number of experiments show that this algorithm can handle light slow change outdoor and light mutation indoor for target de-tection, it also can distinguish the target occluded and target's removal, this algorithm is good accuracy, real-time and robustness.