Online Mulitiple Pedestrians And Vehicles Tracking In Video Surveillance

In the era of information technology, video surveillance has been considerable development, large amounts of data generated from daily monitoring there exists a lot of unnecessary information. These unnecessary information takes up storage resources. How to extract from these videos useful information, it is an important issue now intelligent monitoring systems need to be addressed.Multiple object detection and tracking technology is an important problem in computer vision research. The technology uses computer vision-related technology and machine learning, pattern recognition technology to analyze video data and extract critical information, enabling target detection and tracking.Online multiple object tracking and object detection has a wide range of application scenarios, such as video surveillance, robotics and risk behaviors, etc., which require real time position of the target. This paper proposes a multiple object tracking method based on automatically detected and set up the appropriate practical system. We combine the popular ACF object detection algorithms in industry field and background subtraction method to improved moving object detection results, and we draw on existing methods target tracking formalized as a Markov decision process. For small objects, optical flow tracking is not enough stability and staggered under the object tracking, we made improvements. For small objects, this paper detected foreground regions with background subtraction algorithm, and then the smaller areas considered to be small objects, as a complement to ACF object detection; uniform sampling point for the optical flow may cause optical flow unstablity. So we increase the number of optical flow points in foreground region, reducing the number of optical flow points to improve tracking stability; when people interleave there may have a question that identification assigned wrong. So we use the motion prediction and check the objcet area if change largely. Based on this method, we can reduce the wrong transformation situatoina. Based on these improvements, we have successfully built a robust online multiple object detection and tracking system, and on the MOT and other data sets to test and prove the effectiveness of the verification system.