运用YOLO(You Only Look Once)实时目标检测算法解决了驾驶视频目标检测问题。针对目标检测算法受环境条件影响鲁棒性差、小目标识别能力不高的问题,建立了涵盖多种天气环境、包含疑难目标的驾驶视频样本数据库,提出了疑难样本训练方法...运用YOLO(You Only Look Once)实时目标检测算法解决了驾驶视频目标检测问题。针对目标检测算法受环境条件影响鲁棒性差、小目标识别能力不高的问题,建立了涵盖多种天气环境、包含疑难目标的驾驶视频样本数据库,提出了疑难样本训练方法,训练出可在多种天气环境中良好识别小型汽车、行人、公交车及货车的YOLO检测模型。实验结果表明,该训练方法可有效提升目标检测性能;所得检测模型具有较高的召回率和精确度,可初步应用于实时驾驶视频的目标检测。展开更多
Left-turning traffic without a protected left-turn signal is one of the major safety concerns at urban intersections. Though an average of only l0% - 15% of all approaching traffic turns left, significantly a large pr...Left-turning traffic without a protected left-turn signal is one of the major safety concerns at urban intersections. Though an average of only l0% - 15% of all approaching traffic turns left, significantly a large proportion of left-turn crashes occur involving 21% of all intersection fatal crashes. Where traditional safety countermeasures of signal timing-phasing and use of flashing yellow light have reportedly failed to significantly reduce the rate of crashes, an in-vehicle advance collision warning message can be helpful to reduce left-turn collisions at intersections. In this study, an in-vehicle audio warning application has been designed by providing two safety warning messages (Advance Warning Message and Safe Left-turn Maneuver Message) under the vehicle to vehicle (V2V) communication system, which is triggered based on the acceptable gaps of oncoming opposing vehicles for a safe left-turn. A driving simulator test has been conducted with 30 participants to investigate the impacts of warning messages on performance measures such as speed and acceleration profiles, collision records, brake reaction distance, and intersection clearance time. Statistical results showed that with the help of these messages, all participants were able to reduce speeds and accelerations and chose suitable gaps without potential conflicts. Moreover, the results of questionnaire analysis provide a positive acceptability especially for the Safe Left-turn Maneuver Message. Based on the performance measurements, this type of safety warning messages can be recommended for possible real-road tests for practical applications.展开更多
文摘运用YOLO(You Only Look Once)实时目标检测算法解决了驾驶视频目标检测问题。针对目标检测算法受环境条件影响鲁棒性差、小目标识别能力不高的问题,建立了涵盖多种天气环境、包含疑难目标的驾驶视频样本数据库,提出了疑难样本训练方法,训练出可在多种天气环境中良好识别小型汽车、行人、公交车及货车的YOLO检测模型。实验结果表明,该训练方法可有效提升目标检测性能;所得检测模型具有较高的召回率和精确度,可初步应用于实时驾驶视频的目标检测。
文摘Left-turning traffic without a protected left-turn signal is one of the major safety concerns at urban intersections. Though an average of only l0% - 15% of all approaching traffic turns left, significantly a large proportion of left-turn crashes occur involving 21% of all intersection fatal crashes. Where traditional safety countermeasures of signal timing-phasing and use of flashing yellow light have reportedly failed to significantly reduce the rate of crashes, an in-vehicle advance collision warning message can be helpful to reduce left-turn collisions at intersections. In this study, an in-vehicle audio warning application has been designed by providing two safety warning messages (Advance Warning Message and Safe Left-turn Maneuver Message) under the vehicle to vehicle (V2V) communication system, which is triggered based on the acceptable gaps of oncoming opposing vehicles for a safe left-turn. A driving simulator test has been conducted with 30 participants to investigate the impacts of warning messages on performance measures such as speed and acceleration profiles, collision records, brake reaction distance, and intersection clearance time. Statistical results showed that with the help of these messages, all participants were able to reduce speeds and accelerations and chose suitable gaps without potential conflicts. Moreover, the results of questionnaire analysis provide a positive acceptability especially for the Safe Left-turn Maneuver Message. Based on the performance measurements, this type of safety warning messages can be recommended for possible real-road tests for practical applications.
