Vehicle collision identification method based on optical flow signal

The invention provides a vehicle collision identification method based on an optical flow signal, and the method comprises the following steps: S1, collecting accident-happening and non-accident driving video data public on a network platform, and marking the driving video data; s2, preprocessing all the marked driving video data, and converting the preprocessed driving video data into an optical flow vector; s3, constructing a training sample according to the obtained optical flow vector, and performing model training on the training sample through a deep learning neural network model to obtain a Transform dichotomy neural network model; and judging whether the vehicle is collided or not through a deep learning classification algorithm in combination with the collision video. Collision recognition can be achieved without an external sensor, the cost is low, the accuracy rate is high, and meanwhile a foundation is laid for a main engine plant to achieve accident after-sale care. 本发明提供了一种基于光流信号的车辆碰撞识别方法，包括以下步骤