摘要
为了解决计算机视觉在夜间低亮度、多干扰的光环境下交通流检测精度低、稳定性差等问题,研究了一种基于车头灯光斑特征稳定性与轨迹相似度的夜间交通流量检测方法。针对静态检测方法难以在多干扰的条件下高精度地识别车灯的问题,提出了在跟踪过程中根据目标跟踪窗口的亮度与几何特征稳定性动态消除干扰光斑的车灯识别方法;针对静态配对方法稳定性较差、计算开销较大的问题,提出基于车灯跟踪轨迹相似度与位置关系的配对规则、辅以配对逻辑来实现车灯配对的方法。实验结果表明,该算法在常见检测环境下的正检率约为90%,漏检率可低于10%。算法在应对路面反射特性较强的环境时的漏检率有待优化。
Restricted by low brightness and multi-interference in nighttime,detection of traffic volume using computer vision has low accuracy and robustness.An algorithm is proposed for detecting vehicles under complex traffic conditions in nighttime based on stability and trajectory similarity of headlight.In order to resolve the low accuracy problem of traditional static methods,a dynamic filtering method is used to recognize and remove non-headlight blobs based on brightness and geometry stability of tracking points.In order to resolve the low stability and high computing cost problem of static paring methods,combining paring rules with simple paring logic based on similarity of trajectories and positional relationship of tracking points.The results show that under normal traffic conditions,the positive rate of the proposed algorithm is about 90%;and the false negative rate is below 10%.However,the false negative rate needed to be optimized when reflection from road surface is strong.
作者
林培群
黄子敬
陈丽甜
LIN Peiqun;HUANG Zijing;CHEN Litian(School of Civil Engineering and Transportation, South China University of Technology,Guangzhou 510640,China)
出处
《交通信息与安全》
CSCD
北大核心
2018年第3期56-63,共8页
Journal of Transport Information and Safety
基金
国家自然科学基金项目(61572233)
广东省科技计划项目(2016A050502006
2016A030313786)
广东省交通运输厅科技项目(201502062)资助
作者简介
第一作者(通信作者):林培群(1980-),博士,教授.研究方向:交通图像理解、车联网.E-mail:pqlin@scut.edu.cn;黄子敬(1994-),男,硕士研究生;陈丽甜(1993-),女,硕士研究生。
