Vehicle cut-in early warning method based on visual perception

The invention discloses a vehicle cut-in early warning method based on visual perception, and the method comprises the following steps: 1), collecting images, and forming a first sequence image; 2) identifying vehicle area and position information in the first sequence image by using an instance segmentation model to obtain vehicle mask information; 3) fusing the vehicle mask with the original image to form a second sequence image; 4) inputting the second sequence image into the image feature extraction model to obtain sequence image features; and 5) inputting the sequence image features into a lane changing prediction model, and predicting the possibility of vehicle lane changing. According to the invention, the intention of the surrounding vehicle to cut into the lane where the vehicle is located can be accurately predicted, so that an early warning signal can be sent to a driver in advance to effectively reduce the probability of traffic accidents. 本发明公开了一种基于视觉感知的车辆切入预警方法，包括如下步骤：1)采集图像，形成第一序列图像；2)利用实例分割模型识