基于共生概率特征量的行人检测

Pedestrian Detection Based on Co-occurrence Probability Feature

人体目标检测研究是近年来计算机视觉领域的研究热点。针对行人检测中出现的检测精度较低的问题，文中提出了一种有效的行人检测算法。具体而言，选取不同类型的局部特征量HOG与LBP，通过第一段的Real Ada-Boost算法进行特征的筛选，筛选后的特征通过两两配对计算共生概率特征量；最终通过第二段的Real AdaBoost算法将弱识别器转化为强识别器来进行行人检测。实验以OpenCV和VS2010为测试环境，通过与OpenCV自带的算法程序比较得出该算法能更好的检测行人，从而提高了行人检测的准确率与鲁棒性。

The human body target detection research is a research hotspot in the field of computer vision in re- cent years. In view of the poor pedestrian detection accuracy, this paper presents an efficient pedestrian detection algorithm. Different types of local features HOG and LBP are selected and filtered by the first stage Real AdaBoost algorithm, after which the co-occurrence probability features are generated by pairwise. Finally, weak classifiers are transformed into a strong recognizer to detect pedestrians through the second stage of the Real AdaBoost algorithm. Experiment in OpenCV and VS2010 shows that the algorithm can better detect pedestrian and improve the pedestrian detection accuracy and robustness compared with the OpenCV buit-in algorithm.