Moving Object Detection And Tracking In Intelligent Monitoring System

In the thesis, for achieving new algorithms with better real-time and accuracy, anexhaustive study is made about objects detection and tracking. The following is thedetails of the research work:Firstly, in the respect of object detection, this thesis introduces some common objectdetection algrithoms and correlative image processing techniques, and analyzes thebackground difference algorithm based on Gaussian mixture model. Basing on that, themothod for vehicle detection and count based on virtual loop is putted forward. Set upvirtual loop in image sequence, calculate changes of gray level in it, then judge whetherthere are any passing cars according to the changes, and count the number of them. Theexperiment results indicate that the proposed algrithom without any complicatedcalculations, has high accuracy, real-time and anti-interference.Secondly, in the respect of tracking, the thesis summarizes object trackingcontent,introduces the classification of object tracking algrithoms which are all analized, andmakes a exhaustive study on the object tracking algrithom based on particle filter andcamshift.For single object tracking, the thesis proposes an effient tracing algorithm basedon color information and SURFfusion within particle filter framework, which is moreefficent than the ones based on single feature and lays the foundation for further research.Finally, for multi-object tracking, the thesis proposes a method of particle filterobject tracking combining with camshift. As the foundation for efficent object tracking,the first proposed method in the thesis with high speed and accuracy is used to detect andextract objects in this method. And camshift algorithm is used to track objects in the2times area that the particle filter predicts. So it can reduce the number of particles andimproves the calculating speed and accuracy. If the background is similar with targets,just using camshift algorithm to track objects often leads to the tracking window gettinglarger and even the objects getting lost. This proposed method has solved that problems.