Improved Visual Robot Place Recognition of Scan-Context Descriptors by Combining with CNN and SVM

IF 0.9 Q4 ROBOTICS

Journal of Robotics and Mechatronics Pub Date : 2023-12-20 DOI:10.20965/jrm.2023.p1622

Minying Ye, Kanji Tanaka

引用次数: 0

Abstract

Visual place recognition from a 3D laser LiDAR is one of the most active research areas in robotics. Especially, learning and recognition of scene descriptors, such as scan context descriptors that map 3D point clouds to 2D point clouds, is one of the promising research directions. Although the scan-context descriptor has a sufficiently high recognition performance, it is still expensive image data and cannot be handled with low-capacity non-deep models. In this paper, we explore the task of compressing the scan context descriptor model while maintaining its recognition performance. To this end, the proposed approach slightly modifies the off-the-shelf classifier model of convolutional neural networks (CNN) from its basis, by replacing the SoftMax part with a support vector machine (SVM). Experiments with publicly available NCLT dataset validate the effectiveness of the proposed approach.

查看原文本刊更多论文

结合 CNN 和 SVM 改进扫描上下文描述符的视觉机器人位置识别能力

从三维激光激光雷达进行视觉地点识别是机器人学领域最活跃的研究领域之一。尤其是场景描述符的学习和识别，如将三维点云映射到二维点云的扫描上下文描述符，是前景广阔的研究方向之一。虽然扫描上下文描述符具有足够高的识别性能，但它仍然是昂贵的图像数据，无法用低容量的非深度模型来处理。在本文中，我们探讨了在保持扫描上下文描述符识别性能的同时压缩扫描上下文描述符模型的任务。为此，我们提出的方法在卷积神经网络（CNN）的基础上对现成的分类器模型稍作修改，用支持向量机（SVM）取代了 SoftMax 部分。利用公开的 NCLT 数据集进行的实验验证了所提方法的有效性。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of Robotics and Mechatronics ROBOTICS-

CiteScore

2.20

自引率

36.40%

发文量

134

期刊介绍： First published in 1989, the Journal of Robotics and Mechatronics (JRM) has the longest publication history in the world in this field, publishing a total of over 2,000 works exclusively on robotics and mechatronics from the first number. The Journal publishes academic papers, development reports, reviews, letters, notes, and discussions. The JRM is a peer-reviewed journal in fields such as robotics, mechatronics, automation, and system integration. Its editorial board includes wellestablished researchers and engineers in the field from the world over. The scope of the journal includes any and all topics on robotics and mechatronics. As a key technology in robotics and mechatronics, it includes actuator design, motion control, sensor design, sensor fusion, sensor networks, robot vision, audition, mechanism design, robot kinematics and dynamics, mobile robot, path planning, navigation, SLAM, robot hand, manipulator, nano/micro robot, humanoid, service and home robots, universal design, middleware, human-robot interaction, human interface, networked robotics, telerobotics, ubiquitous robot, learning, and intelligence. The scope also includes applications of robotics and automation, and system integrations in the fields of manufacturing, construction, underwater, space, agriculture, sustainability, energy conservation, ecology, rescue, hazardous environments, safety and security, dependability, medical, and welfare.