Seat Belt Intelligent Recognition Technology Based On Video Big Data

The rapid development of the national economy has enhanced residents' purchasing power of automobiles.However,the traffic safety awareness of Chinese residents is still insufficient,and traffic violations are frequent.Eventually,the number of motor vehicle accidents and casualties continues to rise.One of the important reasons is failure to wear seat belts according to regulations.Therefore,accurate seat belt identification is one of the important functions of the intelligent traffic illegal identification system.Aiming at the problem that the driver's seat belt is difficult to detect in the road surveillance video,this thesis improves the traditional image processing-based seat belt detection algorithm and proposes a seat belt detection algorithm based on weak supervision information.Firstly,the improved image processing-based seat belt detection algorithm uses the Darknet-53 network to locate the window and generate the seat belt detection area.Then the Canny operator is used for edges detection after preprocessing.Finally,in order to identify the seat belt,the probability Hough transform is used to extract straight line from edges.The improved seat belt detection algorithm is an end-to-end window detection method and can effectively identify the seat belt with an accuracy rate of 81.54%,which is superior to most of the image processing-based seat belt detection algorithms.Different from the traditional seat belt detection method,the new method proposed in this paper makes full use of the weak supervisory information to mine the more discriminate features in the image.The algorithm first uses the pre-trained Res Net50 network to extract the global features of the image,then calculates the gradient map of the global feature,selects the important region according to the maximum gradient,and then uses the pre-trained Res Net50 network to extract local information.Finally,the XGBoost classification model is used to recognize the seat belt,which is trained through the reconstructed feature vector.The reconstructed feature vector is consisted of global information and local information.The experimental results show 86.5% accuracy of this algorithm on the surveillance image dataset.Compared with the traditional image processing-based seat belt detection algorithm,the new method proposed in this paper is better.