关键词: 交通事件, 浮动车, 特征分析, 离群检测

Abstract: Traffic incident detection and confirmation is the primary problem in traffic incident management, but detection method based on loop detector and video data is restricted in practical applications due to the high cost and inefficiency. This paper proposes a traffic incident detection algorithm based on outlier mining by use of feature vector related to traffic incident. The feature vector is obtained from the traffic information processed by the float car technology. The algorithm is simple, efficient, and easy to deploy. The experimental results show that the algorithm has higher accuracy than the pattern recognition method, and can effectively distinguish between the conventional congestion and traffic incident.

Key words: traffic incident, floating car data, feature analysis, outlier mining