基于行人关键点的低亮度行人轨迹和意图估计  

Pedestrian trajectory and intention estimation under low brightness based on human key points

作　　者：汪威 罗石[1] 耿国庆[1] 刘军[1] WANG Wei;LUO Shi;GENG Guoqing;LIU Jun(School of Automotive and Traffic Engineering,Jiangsu University,Zhenjiang,Jiangsu 212013,China)

机构地区：[1]江苏大学汽车与交通工程学院,江苏镇江212013

出　　处：《江苏大学学报（自然科学版）》2022年第4期400-406,共7页Journal of Jiangsu University：Natural Science Edition

基　　金：国家自然科学基金资助项目(51675235)。

摘　　要：夜晚行车环境比白天复杂,所以辅助驾驶系统对夜晚行车环境的感知尤为重要.对于行人识别而言,传统的行人轨迹和意图估计算法在夜间的适用性不足.为了能够在夜间估计行人的意图和轨迹,提出一种低亮度环境下近距离行人的意图和轨迹估计方法.使用新的图像增强算法以提高人脸特征识别率.对于增强后的图像,基于YOLOv3和Openpose提出匹配2类结果的算法.进一步采用卡尔曼滤波修正YOLOv3识别误差,再基于匹配结果提出一种行人面部朝向估计算法,然后在上述2点基础上提出估计行人意图新方法.结果表明:图像经过新增强算法处理后,提出的行人轨迹和意图估计方法对于低亮度环境有较好的适用性.The driving environment at night is more complex than that during the day,so the perception of the driving environment at night by the assistant driving system is very important.For pedestrian recognition,the traditional pedestrian trajectory and intention estimation algorithm is not suitable at night.To estimate pedestrian intention and trajectory at night,a method was proposed for estimating pedestrian intention and trajectory in short distance under low brightness environment.The new image enhancement algorithm was used to improve the face recognition rate.For the enhanced image,the matching algorithm was proposed based on YOLOv3 and Openpose.The Kalman filter was used to correct the YOLOv3 recognition error,and a pedestrian face orientation estimation algorithm was proposed based on the matching results to obtain a new method for estimating pedestrian intention.The results show that the proposed pedestrian trajectory and intention estimation method is suitable for low luminance environment after image enhancement.

关 键 词：近距离行人 低亮度图像增强 改进MSR算法 YOLOv3 Openpose 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]