Stopped Vehicle And Abandoned Item Identification Algorithm Research Of Event Detection

Video-based traffic incident detection technology has been a hot spot of global research,it is primarily used to detect some unusual events on the road, such as vehicle retrograde,continuous cross road driving, stopped vehicles, abandoned items, pedestrians. Application ofthe system allows the traffic management department to manage the daily traffic effectively,handle unusual event timely, so it avoid a lot of accidents. This paper focuses on the researchof stopped vehicles and abandoned items detection algorithm.In this paper, the abnormal areas were detected firstly. The first part introduces theimage pre-processing, including image enhancement, image edge extraction, image denoising.In the second part, the abnormal blocks were detected, those blocks have big difference withcurrent image texture in a period of time, they are also the abnormal blocks of stoppedvehicles and abandoned items, this paper detect the abnormal area use the difference ofsteady state. In the third part, the abnormal areas were located, use the mathematicalmorphology processing, and use connected domain to determine the location of the abnormalarea.In order to segment the abnormal area accurately, this paper use Markov random field tosegment local abnormal region, Markov is a method based on the theory of statistics, and alsoin consider of the partial correlation between pixels, make full use of a priori knowledge. Thispaper describes the theory of Markov and modeling approach of characteristic field and labelfield, on this basis, use ICM or simulated annealing to optimize the energy of Markov randomfield, finally, the segmentation of abnormal areas were been obtained.In order to distinguish the stopped vehicles event from abandoned items event,three-dimensional information was been used. Two methods have been proposed, one isbased on3D models, the other is based on the height of three-dimensional. The first methoduse the matched-degree of two-dimensional projection of vehicles’3D model and intersectedareas to distinguish stopped vehicles from abandoned items, high match-degree representstopped vehicles, otherwise represent abandoned items. The second method first obtain the mapping table through camera calibration, then calculate the speed of the different angularpoint in intersected area, the relationship between speed and height can be determined fromcamera imaging model, then the relative height of abnormal region can be obtained, theheight was been used to distinguish stopped vehicles from abandoned items.The algorithm designed in this paper has been tested in a number of different trafficscenarios, the result shows that it can detect stopped vehicles and abandoned items accidents,has high stability, high accuracy, strong practicality and strong robustness. At the same time,it can meet the requirement of real-time detection system.