Real-time vehicle detection using histograms of oriented gradients and AdaBoost classification

Abstract

Vehicle detection is a major part of a driver assistant system. However, a complex environment and diverse types of vehicles make real-time detection of vehicles a very challenging task. This paper proposes a real-time vehicle detection system of two steps: hypothesis generation and hypothesis verification. In the first step, potential vehicles are detected using shadows under vehicles. In the second step, hypotheses generated in the first step are classified as vehicles and non-vehicles. The novel aspect of this research is in constructing two types of histogram orientation gradients descriptors to extract vehicle features, and then combining them for their final features. The AdaBoost classifier is trained by the combined histogram orientation gradients features. The Treatment Group of images vehicle dataset is adopted for the classifier training. The experiment results show that the proposed system performs well in accuracy and robustness and can meet real-time requirements.