基于手势交互的移动机器人三维环境探索及感知技术研究

发布时间：2018-03-25 15:28

本文选题：物体识别　切入点：SLAM　出处：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着机器人技术的进步,移动服务机器人的需求也越来越大。移动机器人的环境感知技术也越来越成为研究热点。作为环境感知的关键技术SLAM技术,近些年也有了很大的进步,机器人的实时定位与地图构建能力得到大幅提升。但是移动机器人在未知环境中首次自主探索的效率低下,构建地图鲁棒性仍然较差,地图构建精度较低,地图的后期使用范围有限。本文以构建实用性较强的机器人环境感知系统为导向,研究了环境引导探索方法、地图构建和地图智能语义标注。为了解决机器人自主环境探索效率低的问题,提出了一种自然人机交互系统来引导机器人完成环境探索。自然人机交互可以有效缩减使用机器人的学习成本,让用户根据已有的经验完成对机器人的操作使用。使用虚拟现实技术作为视觉反馈,使用手势交互作为控制方法,来实现人与机器人的自然交互。在机器人进行环境探索的同时需要完成机器人对环境的地图构建任务。为了能够获取更详细的环境信息,本课题将RGBD SLAM和ORB SLAM进行融合,实现了稠密三维点云地图构建算法。为了让机器人能够真正理解环境信息,本课题提出了一种物体分割定位和特征提取算法。算法使用晶格分析的方法来对空间中的云点进行分析来获取可能的物体分布进而实现物体的分割。在分割完成后根据晶格信息完成物体的定位和特征描述。物体识别算法借鉴词袋模型对物体进行识别,通过物体识别算法即可获得物体的类别信息。将物体的类别信息和定位信息绑定构成语义索引,即可完成语义地图的构建。最后本文构建了完整的实验系统,验证了自然人机交互系统的可行性和易用性。使用稠密三维点云地图构建算法构建了环境地图,并在此基础上构建了物体样本集和训练集,验证了物体分割定位算法和识别算法的有效性和准确率。
[Abstract]:With the development of robot technology, the demand of mobile service robot is more and more great. The environment sensing technology of mobile robot has become a hot research topic. As a key technology of environment perception, SLAM technology has made great progress in recent years. The real-time localization and map construction ability of the robot has been greatly improved, but the efficiency of the first autonomous exploration of the mobile robot in the unknown environment is low, the robustness of the map construction is still poor, and the map construction accuracy is low. The use of map in the later stage is limited. In this paper, the exploration method of environment guidance is studied, which is guided by the construction of a practical robot environment perception system. Map construction and map intelligent semantic annotation. In order to solve the problem of low efficiency of robot autonomous environment exploration, A natural human-computer interaction system is proposed to guide the robot to complete the environmental exploration. The natural human-computer interaction can effectively reduce the learning cost of using the robot. Using virtual reality technology as visual feedback and gesture interaction as control method. In order to realize the natural interaction between human and robot, we need to complete the mapping task of robot to environment while exploring the environment. In order to obtain more detailed environment information, we combine RGBD SLAM and ORB SLAM. The algorithm of constructing dense 3D point cloud map is implemented. In order to make the robot really understand the environment information, In this paper, an algorithm of object segmentation and feature extraction is proposed. The algorithm uses lattice analysis to analyze cloud points in space to obtain possible object distribution and to achieve object segmentation. Then the object location and feature description are completed according to the lattice information. The object recognition algorithm uses the word bag model to identify the object. The object classification information can be obtained by the object recognition algorithm. The semantic index can be constructed by binding the object category information and the location information to complete the construction of the semantic map. Finally, a complete experimental system is constructed in this paper. The feasibility and ease of use of the natural human-computer interaction system are verified. The environment map is constructed by using dense 3D point cloud map construction algorithm, and the object sample set and training set are constructed on the basis of this algorithm. The validity and accuracy of object segmentation algorithm and recognition algorithm are verified.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】