基于YOLOP的车道线检测、目标检测及可行驶区域分割算法部署

Deployment of Lane Detection,Object Detection,and Drivable Area Segmentation Algorithms Based on YOLOP

目前,全景自动驾驶感知中的多任务学习取得了显著成果。其中对象检测和分割任务极其重要,可以帮助决策、提供路线规划和安全信息。然而,目标检测和分割仍有限制,需要大量数据和先验信息。为了使自动驾驶中的多任务学习更加高效准确,本文将嵌入式平台融入车道线检测的研究当中,融合了车道线检测和障碍物检测,可以有效提高计算效率和环境感知。同时,本文使用NVIDIA嵌入式平台对深度学习进行部署并能够保持先进性能。

This paper points out that,at present,multi-task learning in panoramic autonomous driving perception has made significant achievements and achieved remarkable results.Among them,object detection and segmentation tasks are extremely important,which can help in decision-making,route planning and safety information.However,object detection and segmentation still have limitations,requiring large amounts of data and prior information.In order to make multi-task learning in automatic driving more efficient and accurate,this paper integrates the embedded platform into the research of lane detection,integrating lane detection and obstacle detection,which can effectively improve the computing efficiency and environment perception.This paper also deploys deep learning using NVIDIA's embedded platform and maintain advanced performance.