One of the major interest on intelligent transportation systems (ITS) is event recognitions from traffic information gathered by image sensors or spot sensors. For that purpose, the authors employed image sensors due to its more rich information rather than spot sensors. And then, in order to gather precise information from traffic images, the authors have been developing occlusion robust tracking algorithm based on Spatio-Temporal Markov Random Field model. This success has led to the development of an extendable robust event recognition system algorithm based on the Hidden Markov Model (HMM). This system learns various event behavior patterns of each vehicle in the HMM chains and then, using the output from the tracking system, identifies current event chains. By this system, ordinary traffic activities at an intersection were able to be recognized at success rates of 90% average, classifying observation sequences that are obtained from the tracking results. By this system, it becomes possible to classify ordinary traffic activities in detail and abnormal event would be found by distinguishing then from ordinary situations. Actually, the authors could reach a presumable idea of accident detection method. This method recognizes observation sequences similar to HMM model of accidents out of a lot of ordinary traffic activities. And then, combining with recognition results of activities among neighbor vehicles, this system successfully determined pairs of vehicles that are really involved in accidents.
Abstract