驾驶行为分析

对于输入的一张车载驾驶员监控图片（可正常解码，且长宽比适宜），识别图像中是否有人体，若检测到至少1个人体，将目标最大的人体作为驾驶员，进一步识别驾驶员的属性行为，可识别使用手机、抽烟、未系安全带、双手离开方向盘、视线未朝前方、未佩戴口罩、闭眼、打哈欠、低头9种典型行为姿态。

An input in-vehicle driver monitoring image (which can be properly decoded and has an appropriate aspect ratio) supports two-stage recognition tasks: first, detect whether a human body exists in the image. If at least one human body is detected, the human body with the largest target size will be taken as the driver. Then, further recognize the driver's behavioral attributes, which include 9 typical behavioral postures: using a mobile phone, smoking, not wearing a seatbelt, hands off the steering wheel, eyes not looking forward, not wearing a face mask, closed eyes, yawning, and head lowering.

搜集汇总

背景与挑战

背景概述

驾驶行为分析数据集是一个用于识别车载驾驶员监控图片中驾驶员行为的数据集，可识别使用手机、抽烟等9种典型行为姿态。数据集发布于2021年2月5日，每年更新一次，采用增量更新方式。