Obeject Detection Based On Video

As an important part of artificial intelligence,computer vision has been developed rapidly and widely used in recent years with the daily increasing demand.Object detection technology as one of the key technologies of computer vision has become a focus of academic and industrial research in recent years,and achieved remarkable results,and been used in social production and human life.Object detection is the basis of advanced computer vision technologies,such as intelligent video surveillance?object recognition and tracking?behavior analysis and perception,Therefore,the research of object detection has important mean and value.The main content of this paper is divided into two parts that include moving object detection and specific object detection acorrding to the requirements of practical application.Then make an introduction and analysis for the moving object detection and the specific object detection including the research status at home and abroad?the related technology and difficulties?theory knowledge and main methods.According to the main problems,the corresponding solutions are put forward.Firstly,the main methods and their principles of moving object detection are introduced in this paper.The codebook model in background subtraction method is emphatically analyzed,A moving shadow removal algorithm based on codebook model is proposed to remove moving shadows aiming at the problem that the moving shadow will cause the deformation and loss of the detection result,thus the moving object can be detected accurately.Secondly,the methods and principles of deep learning object detection based on candidate region proposal are analyzed.Aiming at the problem that reduce the speed of network model training and detection due to the selective search is slow in the candidate region extraction stage and a large number of overlapping and redundant candidate regions are generated.In this paper,Edge Boxes algorithm is improved by combining visual saliency,and then candidate regions are extracted to train and test Fast R-CNN by Improved Edge Boxes,which improved the training speed?detection speed and detection accuracy of Fast R-CNN effectively.