Novel approach for vehicle detection in dynamic environment based on monocular vision

This paper describes a vehicle detection system in dynamic environment based on monocular vision. Vehicles are separated from forward scenes by the proposed system. Hypotheses extracted using Haar-like feature and Adaboost classifier include several non-vehicle regions. In order to remove false positive detections, we apply the SVM-based classifier with HOG feature and HOG symmetry feature to predicate whether the hypotheses are vehicles in the hypothesis verification process. By this method, false detection rate resulted from only using Haar-like feature is reduced. The vehicle detection system has been evaluated in dynamic environment, and shown a strong and accurate performance.