基于AdaBoost行人检测优化算法的研究

Research of pedestrian detection optimized algorithmic based on AdaBoost

针对车辆辅助安全驾驶系统的道路行人检测和识别问题,对行人检测算法的实时性和检测准确性受光照影响等方面进行了研究。对行人检测识别算法进行了归纳,将在线训练和更新行人分类器技术应用到车辆前方行人检测和识别中。提出了改进的AdaBoost行人检测优化算法,应用该算法行人识别分类器训练时,根据正样本和负样本的错分率,实时调整级联分类器权重系数,在线更新分类器的错误率权重,在保证分类器整体准确率的前提下,降低了分类器的级数,优化了分类器结构,减少了计算的复杂性;采用扩展的类Haar特征,降低了算法对光线的敏感性,提高了环境适应能力。试验结果表明:在相同的检测率下,改进的AdaBoost算法需要的检测时间更少,系统具有更高的鲁棒性,可以满足道路行人识别的要求。

Aiming at the pedestrian detection and identification for vehicle aided safety driving system,a research was conducted which is focus on the problems of real-time and veracity affected by illumination for pedestrian detection algorithmic.The on line updating pedestrian detection classifiers technology was used to detect and recogmze the pedestrian which are in front of the vehicles.An improved AdaBoost pedestrian detection optimum algorithmic was presented.According to positive and negative samples＇ error rate while training pedestrian detection classifiers,the weight of error recognizing rate was on line updated.It could reduce the classifier＇s series and the calculation complexity,further more,guarantee the overall detection rates.The ability to adapting the environment was improved by adopting extended like-Haar characters to reduce the arithmetic sensitivity.The experimental results show that the improved AdaBoost arithmetic need less detecting time and higher robust than traditional method.This arithmetic can satisfy the request of detecting pedestrians appeared ahead of the vehicles.