Generic Video Vehicle Detection Approach Based On Machine Learning

Intelligent Transportation System or ITS, has been proven to be an effective solution to traffic problems. Traffic Data Collection is an important part of ITS, playing a significant role in traffic management and control, transportation planning and many other applications, and it is built upon the detection of road vehicles. Recently, along with the rapid development of computer technology and video image processing techniques, video vehicle detection has become a key technology in ITS. Intelligent video analysis has attracted more and more attentions, and has been widely used in real life.In view of shortcomings of current popular video vehicle detection method based on virtual loops, a universal video vehicle detection approach is presented in this paper, which combines multiple background and vehicle features, and to improve its performance and robustness, machine learning theory is also adopted. The main idea is that, with virtual loops are set in the video image as detection zones, the moving vehicles may cause pixel intensities and local texture changes, and through these changes vehicles can be detected. Multiple pattern classifiers including LDA+ Adaboost, SVM, and Random Forests are used to detect vehicles. Fourteen pattern features, including foreground area, texture change, and luminance and contrast feature, to train pattern classifiers. Then the trained classifiers are applied in new detection videos. Experiments have been taken under various complex dynamic environments,and the results testify the validity and robustness of this method.