基于深度学习的道路行人违规面部区域检测方法

Deep Learning Based Facial Area Detection Method for Pedestrian Violations on Roads

行人可能呈现出多种不同的姿态,包括行走、静止、站立或蹲下等,同时不同行人之间的衣着外貌也有差异。此外,行人目标还可能受到周围环境、光照条件、遮挡物等多种因素的影响,使得面部区域的检测变得更为复杂。为此,提出基于深度学习的道路行人违规面部区域检测方法。结合深度学习中的复合残差学习方法和深度卷积网络,对道路行人违规面部图像实行去噪处理。引入双层子空间的概念,提取道路行人违规面部图像特征。应用YCrCb色彩空间,确定道路行人违规面部区域。实验结果表明:研究方法能够准确检测出交通监控中违规行人的面部区域,且检测误差散度偏低。

Pedestrians may exhibit various postures,including walking,stationary,standing,or squatting,and there are also differences in their clothing and appearance among different pedestrians.In addition,pedestrian targets may also be affected by various factors such as surrounding environment,lighting conditions,and obstructions,making facial area detection more complex.To this end,a deep learning based facial area detection method for road pedestrians violating regulations is proposed.Com-bining composite residual learning methods and deep convolutional networks in deep learning,denoising is applied to facial images of pedestrians violating regulations on the road.Introduce the concept of double-layer subspace to extract facial features of road pedestrians violating regulations.Apply the YCrCb color space to determine the facial areas of pedestrians who violate regulations on the road.The experimental results show that the research method can accurately detect the facial areas of non compli-ant pedestrians in traffic monitoring,and the detection error divergence is low.