Research On Improvement Of Customer Satisfaction In Power Supply Bureau Under The Background Of The Electric Power System Reform

As industrial 4.0,intelligent mobile robot important members of the modern city construction,and in the technology of intelligent mobile robot Simultaneous Localization and map building(Simultaneous Localization and Mapping,hereinafter referred to as the SLAM)to solve intelligent robot problem of "where is my" through the observation in the operation of the robot in the environment,to record their own characteristic information for positioning and build the incremental map.Due to the unique advantages of small size and large sensing range of vision sensors,visual SLAM has become a research hotspot in the field of SLAM in recent years.The traditional vision SLAM assumes that the scene information of the environment will not change during the operation of the algorithm.The multi-view geometry problem is constructed by extracting the scene information features to solve the current vision sensor's own pose.In practical application,because the object and light in the scene will change with time,the pose can't get the real solution,so it can't accurately estimate its position.This paper mainly studies the visual SLAM algorithm in the dynamic environment with illumination changes and moving objects.The main research results of this paper include:1)An image pre-processing algorithm was designed for the image contour changes caused by light changes,and image enhancement was carried out for the situations of too bright and too dark images.Through experimental analysis,the algorithm had an improvement effect on the feature extraction and matching of visual SLAM and the detection module of return loop.2)Analysis of the current mainstream deep learning target detection algorithm,based on the analysis of different algorithms test process and test speed,determine suitable for real-time detection of deep learning Yolov3 target detection algorithm,and aims at Yolov3 images taken in the continuous time existing in target detection leak phenomenon,modified exception frame sliding window algorithm is put forward,and there is no leak phenomenon through the algorithm revised.3)Combined with deep learning target detection algorithm ORB-SLAM2 Yolov3 and visual SLAM algorithm,put forward a kind of dynamic environment based on target detection visual SLAM algorithm,the algorithm is determined by dynamic Yolov3 goal,first visual front-end in the feature extraction to eliminate dynamic,the feature information of the target area for preliminary pose,and through the local built figure threads,thread action to reduce impact on the entire visual SLAM system for dynamic target,finally through the depth of the key frames establish three-dimensional point cloud with colour image map,the experiment shows that Compared with orb-slam2 algorithm,this algorithm has better positioning accuracy and map reconstruction accuracy on the common data set and in the dynamic environment.