Research On Semantic Map Of Indoor Environment Based On Visual SLAM Mobile Robot

In recent years,mobile robots have been widely used in disaster relief,industrial services and other directions.Traditional map construction methods are difficult to meet the needs of mobile robots for high-level tasks.Therefore,it is imperative to improve the semantic information of mobile robot map construction and make it more intelligent.Current methods of map construction tend to deal with static problems,but in dynamic environment there are problems of low accuracy and poor real-time performance.Therefore,in this article,the present map construction methods in the dynamic environment of the introduction of optical flow method and semantic object segmentation method the dynamic interference from SLAM system,ensure real-time and improve positioning accuracy in SLAM system in dynamic environment,coupled with octree method,construct convenient for storage and use of semantic grid map.The main work of this paper is as follows:(1)According to the characteristics of the camera model,the internal parameters and distortion parameters of the camera were corrected and calibrated.Then,the mathematical model is established according to the visual SLAM system,and the four key components of visual SLAM are deeply analyzed.(2)The concept of optical flow method is introduced to illustrate the effectiveness of optical flow method in processing dynamic objects.Then compare several common feature extraction algorithms,determine the appropriate algorithm for subsequent experiments,finally through the optical flow method to realize the detection of dynamic objects.(3)In view of the shortcomings of the traditional ORB-SLAM2 method in removing dynamic objects,this paper proposes a method of semantic segmentation and optical flow fusion based on convolutional neural network,and designs a visual SLAM system that can deal with dynamic environment.Based on the improved ORB-SLAM2 system,experiments are carried out using common data sets.Compared with the traditional ORB-SLAM2 system,the improved method has higher accuracy and robustness when dealing with dynamic environments.(4)To solve the problem that the point cloud map constructed by the visual SLAM system is difficult to be used and developed for the second time,the octree method is proposed to transform the point cloud map into raster map.The public data set and the real environment were used to construct the maps before and after removing the dynamic objects,and the map construction effects of Kinect V1 and Kinect V2 cameras were compared.The experiment proved that the Kinect V2 camera has a better image effect in the aspect of map construction.