基于双阈值运动区域分割的AdaBoost行人检测算法

结合单目摄像机静止拍摄的视频序列使用背景差法或AdaBoost算法检测行人时分别存在易受噪声干扰或检测速度慢的问题,提出一种双阈值运动区域分割的AdaBoost快速行人检测算法。首先建立背景帧,利用前景帧与背景帧的差分图像拟合噪声曲线,提取噪声与亮暗运动目标的阈值,消除噪声,分割出运动区域;然后通过AdaBoost学习算法选择少量有效的Haar-like弱矩形特征构造强分类器;最后在运动区域利用强分类器检测是否包含行人。实验结果表明,该方法迅速缩小了检测范围,加快了检测速度,降低了误检率。

AdaBoost pedestrian detection algorithm based on dual-threshold motion area segmentation

In view of the problem that the pedestrian detection in the video sequences from a monocular fixed camera was easily effected by the noise with background difference method or the detection speed was low with the AdaBoost algorithm respectively, this paper proposed a fast AdaBoost pedestrian detection algorithm based on dual-threshold motion area segmentation by combining AdaBoost with background difference. First, it setup background frame and the foreground frame subtracted it to get the differential image. It extracted the two thresholds between the fitted Gauss noise and the bright/dark motion target from the differential image, and then segmented out the moving areas. Second, it selected a small amount of available Haar-like weak rectangle features to integrate a strong classifier by the AdaBoost learning algorithm. Finally, it adopted strong classifier to judge if the moving areas included the pedestrians. Experimental results show that the algorithm not only reduces the detection range compared to the whole image range and accelerates detection, but also decreases the false alarm rate.