Area Difference-based Vehicle Collision Real-time Detection Method

Traffic accidents caused by vehicle collisions have seriously endangered people's lives and property and this situation has caused extensive concern.A large number of research institutions began to study on video-based vehicle collision detection methods,but most of the methods are computationally intensive and cannot meet the requirements of real-time detection.Therefore,the vehicle collision detection method based on the area difference between the positive and negative regions of the motion interaction field was proposed,which can accurately detect the abnormal events such as vehicle collision in real time.The main work is as follows:Firstly,the research background and research significance of the thesis are introduced,and briefly review researching actuality of vehicle collision automatic detection technology at home and abroad.Secondly,the common moving target extraction algorithms and its advantages and disadvantages are discussed.And then the concept of optical flow is introduced,seven common optical flow algorithms are compared by using the average angular error,average endpoint error and optical flow calculation time.Furthermore,the interaction model between vehicles is constructed by using the characteristics of fluid dynamics.The performance evaluation criteria of the motion interaction field is introduced,and the influence of different optical flow methods on the motion interaction field are discussed.In short,in combination with the optical flow quality,optical flow calculation time and the different optical flow method to extract the motion interaction field performance and other factors such as the Farneback optical flow method is selected to extract the vehicle's motion information,and the Gaussian kernel function is used to construct the ideal motion interaction field.Thirdly,the pros and cons of Kimin Yun's two algorithms are compared.Aiming at the edge problem existing in the Kimin Yun's algorithm,it is proposed to solve the problem by expanding the range of motion interaction fields.At the same time,in order to solve the problem of large calculation in Kimin Yun's two algorithms,the toy car is used to simulate the head-on,side,and rear-end collisions in the real traffic scene.Also,the different characteristics of motion interaction field is found under normal traffic and abnormal traffic conditions so that it can be used to detect abnormal events such as vehicle collisions.It uses a new algorithm that uses the absolute value of the area difference between the positive and negative regions of the motion interaction field.Poor optical flow quality may lead to abrupt changes in the number of positive and negative region contours in the motion interaction field and the inability to detect the end time of the vehicle collision,so the automated regression modeling is constructed by using the water surface propagation characteristics on the time axis,it improves vehicle interaction model based on motion interaction field.The experimental results show that the improved algorithm can not only detect the vehicle that have stopped after the collision,but also effectively improve the detection accuracy of the algorithm.Finally,the effectiveness of the algorithm is verified by eight real traffic monitoring video data,which proves that the proposed algorithm can accurately detect road traffic anomalies in complex traffic environment.At the same time,the algorithm is compared with other existing algorithms,which proves that the proposed algorithm can predict and detect vehicle collision events in video with high accuracy.