Key Technologies Research On Three-dimensional Map Building And Optimization Based On 3D Camera

This paper studies the mobile robot three-dimensional(3D) color mapping and localization based on 3D camera, and discusses the key frame extraction and loop closure detection technology, then achieves the optimization of the map. It's main work is as follows:Firstly, this paper choose Kinect 3D camera as the equipment to get 3D visual information, and combines with Pioneer mobile robot to build the hardware equipment. At the same time, uses PCL(Point Cloud Library). So the development equipment of mobile robot 3D mapping and location is built already.Secondly, this paper presents a complete 3D color mapping and localization scheme, which includes the introduction of the description of robot's pose and coordinate transformation. Integrating with the six degrees of freedom parameter model and feature point extraction and match base on ORB(Oriented FAST and Rotated BRIEF) algorithm, the robot's 3D movement estimation is completed based on ICP(Iterative Closest Point) algorithm; On the basis of the robot's initial pose, this thesis establishes the 3D map initialization and update principle, and then displays the map and location information. The theoretical value and the real value of the experimental results are used to prove the reliability of motion estimation and mapping programs. However, the scheme has accumulated error that makes map deformed, so it's need to be further optimized.To solve these problems, this paper proposes an optimization solution based on key frame loop closure detection and global map. The method of loop closure detection based on key frame extraction and Euclidean distance excluded is improved, and a program combining entropy rate, the feature point extraction and the Euclidean distance is put forward innovatively. Experiments both in similar and non-similar environment verify that the key frame extraction technology is robustness. At the base of the previous loop closure detection means, this article innovatively proposed a new way to detect the loop closure based on Euclidean distance elimination and best closed loop selection. To make sure the loop closure detection entirely correct, the paper uses the feature point number as the preliminary detection, and then utilizes inter entropy rate and Euclidean distance as the further confirmatory analysis. Experimental analysis of simple track, complex track and multi-track rail's loop closure detection results, that ensure the detection of stability and consistency.Finally, thesis uses Stochastic Gradient Descent algorithm to optimize the map, after getting results by the improved loop closure detection method. And experiment analysis the indicators before and after the map optimization, the results show that the optimization has improved the performance of 3D mapping and location.