Real-time boosted vehicle detection deal with high detection rate using false alarm eliminating method

We present in this paper a high detection rate of Boosted vehicle detection. The positive database of Boosted training usually consists of similar contour or lighting of vehicle image, but in our work the training data is extracted from both daytime and evening images, it means different in lighting, and we are forced to stop Boosted training toward a high detection rate, which results in a relatively high false alarm rate. Therefore, the Boosted detector will detect vehicle candidate as much as possible. Moreover, we develop two false alarm eliminating methods to eliminate the false vehicle candidates. The algorithm consists of edge complexity for daytime case, and combination histogram matching and intensity complexity for evening case. Each case is chosen by automatic switcher algorithm. To provide real-time detection, we also proposed position based sliding window detector. The idea is based on sliding windows size selection relative to the position of vehicle candidate in an image. Thus, we do not need to apply various size of detector size as traditional Boosted algorithm. Finally, our experimental results show that our proposed system can operate in a real-world environment, while providing realtime detection. © 2013 ICIC International