摘要
针对公交客流统计中因光照、拥挤等引发的统计精度较差问题,提出一种适用于硬件平台的立体视觉客流统计方法及系统。通过优化后的MC-CNN网络获取由双目视觉匹配得到的视差图,处理转化为深度图像去实现乘客头部轮廓的分割、识别、跟踪与计数,有效解决了因遮挡、光线变化等带来的虚假目标和统计精度变差的问题。
In order to solve the problem of poor statistical accuracy caused by illumination and congestion in bus passenger flow statistics,a method and system of stereo vision passenger flow statistics suitable for hardware platform are proposed.Through the optimized MC-CNN network,the parallax map obtained by binocular vision matching is obtained,and the processed image is converted into a depth image to realize the segmentation,recognition,tracking and counting of passenger head contour,thus effec⁃tively solving the problem of false targets and poor statistical accuracy caused by occlusion and light change.
作者
刘彬
魏海峰
张懿
李垣江
LIU Bin;WEI Haifeng;ZHANG Yi;LI Yuanjiang(School of Electrical and Information,Jiangsu University of Science and Technology,Zhenjiang 212003)
出处
《计算机与数字工程》
2023年第3期690-694,711,共6页
Computer & Digital Engineering
关键词
双目视觉
特征提取
立体匹配
公交车客流量
binocular vision
feature extraction
stereo matching
bus passenger flow