摘要
针对目前车辆压线检测系统存在易受环境光照变化影响、拍摄角度固定等问题,本文设计了一种包含图像亮度调整模块和基于深度神经网络分割模块的车载端违章压线检测系统。系统首先利用图像亮度调整方法对图像进行预处理,再利用基于深度神经网络的车辆目标检测和语义分割模型从图像中获取车辆和车道线信息。论文中利用轻量级空间卷积模块加强车道线分割模型特征提取的效率,实现高精度车道线分割,满足实时性要求。最后根据车道线分割结果和车辆目标检测结果进行压线行为检测。实验证明,本系统能够有效检测车辆违章压线行为,同时对环境光照变化具有一定的鲁棒性。
Aiming at the lane-crossing detection system that is susceptible to changes in ambient light and fixed shooting angles,this paper designs a vehicle-mounted lane-crossing detection system that includes an image brightness adjustment module and the deep-neural-network based vehicle detection and lane segmentation module.The system first uses the image brightness adjustment method to preprocess the image,then uses the deep-neural-network based detection and segmentation model to obtain vehicle and lane information from the image.In the paper,a lightweight spatial convolution module is used to enhance the efficiency of feature extraction of the lane segmentation model,achieve high-precision lane segmentation results,and meet real-time requirements.Finally,the lane-crossing behavior is detected according to the result of lane segmentation and vehicle detection.Experiments have proved that this system can effectively detect lane-crossing behavior,and at the same time is robust to changes in ambient light.
作者
于润润
朱凯赢
蒋光好
吴益
周伟
YU Runrun;ZHU Kaiying;JIANG Guanghao;WU Yi;ZHOU Wei(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《智能计算机与应用》
2021年第10期147-150,共4页
Intelligent Computer and Applications
关键词
深度学习
目标检测
语义分割
压线检测
deep learning
object detection
semantic segmentation
lane-crossing detection
作者简介
于润润(1995-),男,硕士研究生,主要研究方向:计算机视觉、语义分割;朱凯赢(1993-),男,硕士研究生,主要研究方向:视觉SLAM;蒋光好(1995-),男,硕士研究生,主要研究方向:行为识别;吴益(1996-),女,硕士研究生,主要研究方向:三维目标检测;周伟(1996-),男,硕士研究生,主要研究方向:视频异常检测。
