交通场景行人穿越检测方法进展研究

Research on Pedestrian Crossing Detection in Traffic Scenes

随着自动驾驶和计算机视觉技术的飞速发展,人们对交通安全的重视程度越来越高。其中,交通场景中行人穿越检测问题是研究的热点与难点之一。对行人穿越检测这一领域进行了综合评述,并对未来研究进行了展望。按照行人试图穿越、正在穿越及穿越预测三个阶段进行分析,首先分析了基于行人姿态穿越意图的研究进展(试图穿越),包括人体姿态、数据驱动和高斯动力学模型三个方面;其次讨论了基于轨迹跟踪的穿越行为检测方法(正在穿越);然后概括了基于轨迹预测的行人穿越检测方法(穿越预测);最后总结了现阶段比较新颖的多传感融合、车路协同、车联网等研究角度,并对未来该领域需重点发展的方向提出建议。

With the rapid development of autonomous driving and computer vision,more and more afforts are being made for traffic safety research,among which pedestrian crossing detection in driving environment is one of the most prominent problems.The field of pedestrian crossing detection is reviewed,and its future research is prospected.According to the three stages of pedestrian crossing process-crossing intention,crossing behavior and crossing prediction,firstly,the crossing intention including three aspects:human posture,data-driven and Gaussian dynamic model is reviewed.Secondly,the tracking frameworks in crossing are discussed.Then the pedestrian crossing prediction works based on trajectory prediction is summarized.Finally,some new research perspectives including multi-sensor fusion,vehicle road cooperation and vehicle Internet are summarized,and some suggestions for the future development of this field are put forward.