自适应人体运动目标精检测

Moving Human Body Fine Detection Based on Self-Adaptive Threshold

现有人体检测算法普遍存在检测精度不足,适应环境能力差的缺点,为了改善这种情况,提出了一种结合区域RGB权值和自适应阈值的人体精检测算法。该算法首先通过背景差分法对前景人体目标进行快速检测,分离出近粗略人体目标区域,然后根据人体的特征将人体可能的区域范围确定,结合两次检测的区域估计出人体目标区域。将目标区域分割为若干大小相同的小块,分别对每一个小块计算RGB权值和检测阈值,并归一化到(0,255)区间,利用加权后的新值和得到的阈值通过背景差分法进行精检测,得到最终结果。实验结果表明:本文的检测算法可以比现在流行的基于HIS的人体检测算法精度提高10%左右,比普通的背景差分法检测精度最高可提高40%左右。通过实际验证,本文的算法完全满足人体检测对精度和适应性的要求。

The existing human body detecting algorithms often have the shortages,such as less precision and less adaptability of environments.In order to improve the detecting results,an optimized algorithm is presented based on the combination of the weight of RGB channel and self-adaptive threshold in sub-area.Firstly,a simply background subtraction is applied to detect the moving human body area cursorily;secondly,the human area is calculated and estimated by its characters,then the two areas are combined to get the final target area.The target area will be divided into several sub-areas with same size,and then the weight of RGB channel and the threshold of each sub-area can be calculated and normalized to the range of ,respectively.By using the calculated threshold and the weight value of target area＇s RGB channel,we can do a fine background subtraction to detect the final result.Experiment results show that the precision of our algorithm has improved by 10% as compared with the algorithm based on HIS,and has improved by 40% as compared with the traditional algorithms.It can wholly achieve the requirements of precision and adaptability by actually testing.