Detection of abnormal vehicle behaviours using traffic videos

On highway traffic flow, drivers move their vehicles to be able to travel safely by complying with certain rules. These rules are determined to maintain the traffic flow without damaging to each other of moving or stationary vehicles and pedestrians on highways. Unless the rules are obeyed by the vehicles or pedestrians, traffic accidents that can cause loss of lives and properties in vehicle or external environment may occur.In this study, design and implementation of a system to detect the abnormal behaviors of vehicles by using videos is done. Firstly, the normal traffic flow is learned by the system. Therefore, trajectories of vehicles which were tracked in the normal traffic flow are clustered using continuous Hidden Markov Model to determine the highway movement patterns.After the determination of the highway movement patterns using the camera captures of normal traffic flow, study was progressed to detect the abnormal behaviors of vehicles. In this section, trajectory based abnormal vehicle behaviors (moving outside of the lane or in the opposite lane, wrong U-turns) and relative velocity based abnormal vehicle behaviors (sudden deceleration or acceleration) are detected. For this purpose, classification of the partial vehicle trajectories between highway movement patterns is done by using the similarity between movement patterns and partial trajectories. Furthermore, sudden unexpected changes in vehicle speeds are also detected as vehicle abnormal behaviors.According to experimental results, abnormal behaviors of vehicles that have accuracy ratio of 85% precision ratio of 87% are detected. Undetected or wrong detected abnormal vehicle behaviors are caused by factors such as uncompleted vehicle tracking and extreme shift of the center point of the vehicle during the vehicle tracking.