Research On The Detection Of Abandoned Objects In Highway Based On Video

With the development of economic and the progress of social, highway has a rapiddevelopment, which makes traffic problems are becoming serious, traffic accidentsoccur frequently. Abandoned objects events on the highway as a frequent traffic events,which triggers traffic accidents and causing serious security hidden danger has becomethe urgent need to resolve problems. On the highway, how to accurately and rapidlydetect occurrence of abandoned objects, and early eliminate potential safety hazard,guarantee the traffic flow and safety, has become a hot issue of concern.In the paper, through analyzing the problem of detecting the abandoned objects onhighway deeply, the issue is divided into three subproblems, and discuss their solutionsrespectively. For the first problem of obtaining reliable background, through analyzingthe three common algorithm for background extraction, and according to the specificscene of the highway, this paper presents a method based on block partition withalgorithms for background extraction and updation, and the experimental results showthat the algorithm can get reliable background; For the second problem of findingaccurate and complete target detection algorithm, after discussing the common threetargets detection algorithm, this paper presents the twice-segmentation target detectionalgorithm, the improved algorithm based on cellular automata and based on the outlineof mass contains fusion algorithm, the three algorithm commonly carry out the targetextraction, experimental results show that the extracted target is more complete andaccurate; For the third problem of finding the algorithm to accurately and rapidly detectabandoned objects, according to the characteristic of abandoned objects on highway andits the general characteristics with the common static foreground thing, this paperpresents a abandoned objects detection algorithm of static pixel-statistical based onblock, and combining the detection algorithm based on cancroids predicted Kalmanmulti-objective tracking and based on the Support Vector Machine (SVM) targetclassification, experimental results show that can detect accurately and rapidly throwingobjects from scene.