Implementation of Vehicle Recognition System Using Affine Rotation Transform Tracking Framework and CNN Algorithm Base on GPU

摘要

In recent years, with the rapid increase in accidents or crimes, they are exposed to danger not only in public places but also in private places. Faced with these problems, we feel the need for security and surveillance systems. Therefore, the demand of intelligent real-time surveillancesystem is increasing, and a lot of technologies such as object tracking, detection and object recognition in a camera for use in an automatic surveillance system are being developed. In this paper, we propose a realtime vehicle monitoring system using CNN (Convolution Neural Network) learningalgorithm. The process is largely a preprocessing, learning and recognition. The preprocessing consists of object tracker using optical flow algorithm, object detector using Ferns algorithm, and affine transformation algorithm to prevent noise by rotation transformation of object informationobtained through it. First, the object of interest in the specified region of interest is detected and tracked through the Ferns algorithm and uses the sparse optical flow algorithm. The Ferns algorithm and the optical flow algorithm for detection and tracking start simultaneously in the preprocessingand maintain a complementary relationship between detection and tracking. In addition, Affine transformation algorithm is used to remove objects such as distortion, obscuration, and rotation. We implement a video system that recognizes and classifies objects by learning the feature vectorclass of the object of interest through learning GPU-based convolution neural network learning algorithm.