Research On The Road Scene Recogniition Of Outdoor Mobile Robot Based On Vision

For the autonomous mobile robot working outdoors,it has the ability to understand the outdoor road environment.In this paper,an experimental platform of mobile robot is built,and the outdoor road scene understanding and vehicle target tracking research are carried out on this platform.The main research contents of this paper are as follows:(1)According to the characteristics of the outdoor environment,a robot platform that can move freely in the outdoor environment is built,and a sensor system composed of camera and lidar is designed on the platform,as well as the corresponding power circuit and drive circuit,providing a hardware basis for the subsequent algorithm research.(2)Aiming at the problem of outdoor environment perception and scene understanding,a method of outdoor road scene understanding based on deep learning semantic segmentation was proposed.The traditional image semantic segmentation algorithm requires manual labeling of features,and the process is complicated and has large limitations.The deep learning semantic segmentation algorithm can solve these problems,and the segmentation effect is better.In this paper,two semantic segmentation algorithms SegNet network and ENet network are studied,and the algorithm is improved and optimized to improve the network speed.It has been proved by experiments that this method is fast,accurate and adaptable,which greatly improves the deep learning reasoning speed on the embedded ARM platform,and can provide correct road information for outdoor robots in real time.(3)According to vehicle targets under the outdoor environment problem,was proposed based on camera and laser radar sensor fusion method,this method is higher than using a single sensor method the accuracy and reliability of the,anti-interference performance is stronger,can be more accurate estimates of the vehicle in front of the mobile robot distance,ensure the safety of the robot.(4)In the outdoor environment,through the experiment on the road recognition and obstacle detection algorithm has made the comparison and analysis,in terms of semantic segmentation algorithm,respectively for real-time semantic network and the improved SegNet ENet network segmentation experiments,through qualitative analysis and quantitative analysis,get the results of the experiment SegNet network segmentation speed and segmentation effect is better,to meet the requirements of the outdoor mobile robot to scene understanding.In the understanding of obstacles in outdoor scenes,the accuracy of vehicle distance estimation by sensor fusion algorithm is more accurate than that by single sensor detection.