Research On Vehicle Blind Spot Detection Method Based On Deep Learning Of Digital Image

At present,the penetration rate of family cars is increasing year by year,and the frequency of traffic accidents is also increasing.With the rapid development of deep learning in the field of artificial intelligence,the intelligence of vehicle safety auxiliary equipment has been rapidly developed.This paper proposes a vehicle blind spot detection system based on digital image processing.The real-time picture of the rear-view monocular camera is used as input,and the object detection algorithm and target tracking algorithm based on deep learning are combined with the distance measurement model to realize the detection of vehicles in the blind spot of the vehicle.Accurate identification,positioning and alarm.In order to ensure that it can be deployed in embedded devices with limited computing power,so that it can be detected quickly and accurately,it is necessary to optimize the above core models or algorithms.This paper mainly does the following.(1)This paper uses the object detection algorithm-YOLOv5 as the core of the detection task,identifies the target vehicle behind through the monocular camera,and generates the detection frame in real time.The improved Bo T-YOLOv5 introduces the Bo T module based on the Transformer mechanism and the CBA module based on the attention mechanism to improve the feature extraction ability and detection ability of the model while compressing the model parameters to ensure that the target can be detected quickly and accurately in the vehicle equipment.(2)This paper designs an automatic calibration algorithm based on Bo T-YOLOv5,and establishes a ranging model based on a monocular camera.First use the Zhang Zhengyou calibration method to calibrate the internal parameters of the camera,and then combine Bo T-YOLOv5 with the K-means clustering algorithm to automatically generate road vanishing points through the lane line bounding box,which is used to calculate the external parameters of the camera,and establish a pixel coordinate system to the world coordinate system The conversion relationship of the pixel plane is used to complete the distance measurement of the target vehicle in the pixel plane.(3)Aiming at the problem of discontinuous vehicle detection due to vehicle occlusion,an improved target tracking algorithm Deep-SORT is proposed,using Bo TYOLOv5 as a detector,and the Kalman filter is used to predict the target by introducing acceleration parameters.For the trajectory of the vehicle,GIOU is used to strengthen the correlation and matching degree between the bounding box and the prediction frame,and the actual trajectory of the target is drawn in combination with the ranging model to achieve precise positioning and tracking.(4)Complete the function design of the vehicle blind spot detection system.According to the established world coordinate system,delineate the range of blind spots on both sides of the rear of the vehicle in the image.After the improved Deep-SORT algorithm obtains the actual trajectory of the target vehicle,carry out Positioning and collision time estimation are used as the judgment basis for conventional blind spot detection early warning and secondary alarm,and the detection program and weight model are deployed to the device to verify the detection effect in actual road conditions.