Vehicle detection under tough conditions using prioritized feature extraction with shadow recognition

Abstract

Vehicle detection is the core function in any Driver Assistant System. Besides the challenge in various environmental conditions, the limitation in execution time and computing power is also critical. This paper proposes a shadow detection step that aims at recognizing the shadow part of the train in various environments (including very tough cases) to accelerate the detection process. We propose two shadow recognition approaches for railway trains. In our first approach, we propose a prioritized feature extraction scheme that examines multiple features such as HOG and Color Histogram hierarchically to achieve high robustness as well as preserve the fast detecting speed. Experiments show satisfying results. Subsequently we propose a second approach using machine learning that automatically learns the features and decisions via a modified decision tree classifier with a novel confidence measuring scheme. Experiments show further improvements in both accuracy and execution time.