Research On Moving Object Detection And Behavior Recognition Technology In Recording And Broadcasting System

Video recording and broadcasting system get extensive valuing along with the development of computer network technology,and a lot of research are expanding.As the technical core of video recording and broadcasting system,video-based moving object detection and behavior recognition can improve teaching quality and automation degree of video recording and broadcasting system,so which is applied widely.In accordance with the demand of recording and broadcasting system,moving object detection and behavior recognition technology are researched,and a moving object detection and behavior recognition method based on three frame difference and improved motion history image is designed in this paper.The accuracy of the proposed method can approach more than 92%,and the processing rate can reach 30 frames per second.In term of moving object detection,the paper analyzes and simulates mean filter,Gaussian filter and median filter of image preprocessing stage at first,and select median filter as the image preprocessing method.To overcome the deficiency of high computational complexity in median filter,a fast calculation method of median filter algorithm is proposed.Then three frame difference take the place of frame difference,according to the ghosting and stretching caused by the frame difference method in the target detection.The common threshold segmentation method after difference is studied and compared,and maximum inter class variance method is selected in three frame difference.Finally,the combination form of the basic operations of morphology is confirmed through simulation experiments,in order to make the target obtained by three frame difference method more connected.For behavior recognition,motion history image improved with regional optical flow to resolve the defect that it is difficult to reflect the local motion information of single frame target in motion history image.The improved motion history images can distinguish the similar movements better.Then,we compare Hu moments and Wavelet moments,and the wavelet moments which can reflect the details of the image are selected to extract the features of the improved motion history image.Finally,the K nearest neighbor and support vector machine classification algorithm are compared,and this article choose support vector machine with more ideal classification effect to classify the features.Finally,the appropriate hardware platform is chosen,and target detection and behavior recognition method proposed in this paper are implemented on the hardware platform.By testing in the real environment,the proposed method has good performance in accuracy and real-time,it can meet the practical application requirements.