摘要
针对传统的人头检测算法很难解决公交车上乘客拥挤、穿戴衣帽复杂及光照变化对客流量统计造成影响的问题,本文提出一种新的结合梯度分层的最大稳定极值区域(maximally stableextreme regions,MSER)检测算法。首先对深度图像进行形态学处理,使用结合梯度分层的MSER进行分割,利用面积与深度均值关系对区域点集进行筛选,最后利用圆形标识对符合条件的人头区域进行标记,并将该算法在公交车上进行测试实验。实验结果表明,该算法检测识别率达到98%以上,有效解决了人群拥挤、车内光照变化等诸多问题对统计的影响。该研究满足实时性要求,具有一定的应用价值。
It is difficult to solve the push of passengers, complex clothings, illumination changes, and so on by using the traditional head detection algorithm. Due to this problem, a novel MSER detection algorithm combi-ning with gradient hierarchies is proposedin this paper. Firstly, the depth images are morphologically processed and segmented using the Maximized Stable Extreme Regions, which are combined with the gradient. Then, the regional point set is screened by using the relationship between area and depth mean. Finally, the eligible head areas are marked as the circular marks. After a lot of experiments on the bus, the result shows that the head recognition rate based on the proposed algorithm is more than 98% , the algorithm accurately extracts the head area of the passenger, which effectively solves the problems, such as crowd, mutual occlusion, complex cloth-ings and illumination changes. The algorithm satisfies the real-time requirement and has certain application val-ue.
出处
《青岛大学学报(工程技术版)》
CAS
2017年第4期42-45,53,共5页
Journal of Qingdao University(Engineering & Technology Edition)
基金
国家自然科学基金资助项目(61I70106
61305045)
山东省科技发展计划资助项目(2014GGX101048)