Rapid vehicle detection method based on convolutional neural network

The invention discloses a rapid vehicle detection method based on a convolutional neural network. The method comprises the steps of dividing a to-be-detected image im0 into a plurality of pictures with different scales according to a fish eye imaging principle for detection; 1) building a mobilenet _ yolov3 model on a pytorch; 2) preprocessing the image, intercepting and scaling the to-be-detected image to obtain three images, namely im1, im2 and im3 respectively; 3) respectively detecting im1, im2 and im3 by using three threads; and 4) restoring the detection results of the three pictures into the coordinates in the original im0. According to the invention, the deep learning is landed in an embedded adas system based on an rk3399 platform, the multi-thread detection time is shorter than 120 ms, the vehicle is tracked after detection, the practical application can be achieved, the image detection is carried out in a targeted manner according to the actual road conditions in the adas system that the distant ta