Research On Indoor Environment Perception Method Based On Deep Learning

Environmental perception is an important research direction in the field of computer vision,which is the basis for realizing the visual task of understanding various scenes and is the key to realizing the intelligence of indoor service robots.And the technology related to object detection,binocular ranging and environmental perception is the research focus of this paper.The application of this technology can help indoor robots obtain the position,category and distance information of target objects in the indoor environment,and help them to better perform one step.At the same time,in order to make indoor service robots better complete the identification and distance measurement of some common indoor targets,and considering that some smaller service robots have weak computing power of hardware equipment.Combining the actual application requirements of indoor service robots,the article conducts in-depth research on object detection technology and binocular stereo vision technology for the application of object detection technology and binocular stereo vision technology in environmental perception,and completes the light-weight environment perception system,to achieve object detection and ranging functions,the main functions of this paper are as follows:(1)For the environmental perception of service robots in indoor environments,combined with the needs of real-time detection and ranging,and the operating efficiency of existing experimental equipment,this paper takes lightweight target detection as the research purpose,and selects lightweight target detection algorithms as the initial Study algorithms.At the same time,considering the diversity of indoor targets and the accuracy of detection,this paper extracts several common indoor targets from the public data set COCO2017,and makes them into a new data set as the research data of this paper,and completes the target detection model.training and evaluation.(2)For the target detection method,this function is optimized based on the YOLOv5 s algorithm to realize a lightweight target detection system and ensure detection accuracy.In this paper,the backbone network,feature fusion layer and loss function are improved in three parts to complete the establishment of the new model.Use the data set in this paper to complete the training and evaluation of the improved target detection model.At the same time,the improved algorithm is verified in different data sets and occasions.According to the experimental results,the improved model achieves a balance between light weight and precision,and can meet the real-time and precision requirements of target detection.(3)For the binocular ranging method,this paper adopts the target ranging method based on binocular stereo vision.Based on the imaging principle of the camera and the basic theory of binocular stereo vision,combined with the actual needs of this paper and the selection of existing equipment,the calibration of the binocular camera is completed using Zhang Zhengyou’s calibration method.After the calibration is completed,use the obtained internal and external parameters of the camera to complete the stereo correction of the binocular camera image to meet the needs of subsequent stereo matching.The local stereo matching BM algorithm and the semiglobal stereo matching SGBM algorithm are compared through experiments,and the SGBM algorithm is selected to obtain the disparity map.Finally,the target distance is estimated according to the binocular ranging formula,so as to obtain the distance information of the object in the real scene.The experimental results show that at short distances,the target ranging method based on binocular vision has a high measurement accuracy.(4)For the fusion method of target detection algorithm and binocular ranging method,establish an environment perception system to obtain target position,category and distance information,and complete the experiment on the basis of existing equipment.Simulate the working scene of the indoor service robot,use the binocular camera to acquire the scene,and use the improved target detection model to complete the target detection,and use the fusion method of target detection and binocular ranging to complete the target detection and ranging.Experimental results show that the model has good detection performance and can obtain accurate target distance information for targets in indoor environments captured by binocular cameras.