Recognition And Mapping Reconstruction For Robot Based On Distributed Modular Technology

In order associate 3D map with indoor and semantic information, it requires indoor mobile robot not only to establish a secure environment map, while classifies the scene. The large computing complexity which key technologies of 3D mapping, object classification had, has restricted the development of the robot. To solve these problems, this article implements a simultaneous recognition and 3D mapping robot system based on distributed modular technology. In this system, the paper in RGB-D information, uses image pixel local eight connected structure, and fuses depth-first algorithm to optimize the original depth map. we use RANSIC improved ICP to estimate robot’s position and reconstruct 3D map, while utilize convolutional neural network to classify objects in key frame. We improve the structure of the overall system with distributed modular technology. Timeliness and flexibility of the system has improved significantly. In order to synchronize the two processes, we also propose a method of synchronizing identity.The main work of this paper is as follows:Convolutional Neural Network deep learning model is introduced to realize the scene recognition and classification of the robot.Aiming at the object recognition problem of robot, this paper introduces the CNN deep learning model, building 8 layers depth convolutional neural networks in the recognition process, including 5 layer coil layer and 3 layer fully connected layer. The Caffe architecture, and through the Image Net data for training and the CUDA accelerated algorithm to achieve real-time object recognition and classification.To realize 3D construction technology based on RGB-DThe Kinect obtain scenes color and depth images, combined with the depth-first algorithm for filtering the original depth map, and use ICP iterative closest point algorithm and RANSIC algorithm, match the three-dimensional information of the multi frame color image feature points, realize pose estimation of the robot, update pose estimation in order to reduce the error of dense point clouds accumulated with key frame information, finally to complete three-dimensional reconstruction of the scene of the robot.Improve overall system structure by using distributed modular technologyWe use distributed modular technology to parallel the recognition and reconstruction processes, and propose a labeling method for these processes synchronization. The real-time character of the system is improved, and the distributed modular technology enables system structure more flexible. The increase of The functional component reusability is improved, and allowing the robot to have advantages on cross platform and programming language compatibility.Use pioneer3 DX as experimental robot platform, experiment the distributed modular robot system in really environment in the laboratory. Experimental results verify the presented feasible and effective of the scene recognition and reconstruction method in this paper based on the distributed modular technology of robot.