Research On Visual SLAM System Of Home Service Robot

SLAM(Simultaneous Localization and Mapping)is an important research topic for indoor positioning and navigation technology.SLAM enables mobile robots to locate themselves in real-time and reconstruct incremental environmental maps in the process of movements in unknown environments.Visual SLAM is an important branch of SLAM technology.It has the characteristics of low sensor price and rich map information,so it has become the research direction of SLAM which has been rapidly emerging in recent years.But in fact,the research of Visual SLAM technology is still in the preliminary exploration stage.When it is applied to the actual scene,there will be many problems such as poor environmental robustness and low positioning accuracy.Based on the above situation,this paper studies the Visual SLAM system of home social robots in the context of the family environment.1)The paper analyzes and compares the performance of visual odometer based on feature method and direct method,and gives suggestions for use.We first introduce the principles and differences of the direct method and the feature point method.Then,based on the frequent changes of illumination in the home environment,and considering the current computing power of the mobile platform is limited,we designed two experimental indicators of illumination variation robustness and pose estimation time overhead,and compared the visual odometers of the two methods.The experimental results show that the pose estimation speed of the direct method is several times of the feature point method,while in terms of the robustness of illumination variation,the performance of the feature method is obviously better than the direct method.Finally,based on the experimental results,we give suggestions for the selection of SLAM methods under different working environments and platforms.2)The paper proposes an improved ORB-SLAM2 algorithm.Firstly,a binary-based vocabulary storage method and a vocabulary training algorithm based on the improved ORB operator are proposed improve the system startup speed and tracking positioning accuracy,and reduce the vocabulary size and lightweight system.Then,the offline map construction method and the robotic fast relocation algorithm based on offline map are proposed,which enables the system to realize map reuse,thereby greatly improving work efficiency.Finally,we propose an offline visualization method for map elements and mapping trajectory,so that the system can obtain better human-computer interaction performance.3)In this paper,the proposed algorithms are implemented,integrated and tested,and the results are analyzed.We built the social robot platform and implemented the algorithms using C++ programming language.The experimental results show that the home social robot Visual SLAM system based on improved ORB-SLAM2 can reuse offline map to achieve fast relocation.It has good performance in terms of starting response speed,tracking and positioning accuracy,and human-computer interaction.It is also robust to dynamic changes in the environment.