摘要
针对在摄像机视角、光照、气候、地貌等条件的大幅度变化或者存在快速移动物体的复杂场景下,视觉即时定位与地图构建(simultaneous localization and mapping,SLAM)的精确性和鲁棒性较低等问题,闭环检测作为解决SLAM位姿漂移的重要环节,提出了一种基于神经网络的闭环检测方法。该方法通过传感器获取视觉图像的数据,不同于传统方法的特征提取,采用改进三重约束损失函数训练Darknet提取图像特征,构造对应特征向量矩阵。由于Darknet借鉴了残差网络(resnet)的思想,在具有较深网络层数的同时,仍保持较高的准确率,减少了特征提取误差。经过自编码器方法对数据进行降维处理,通过余弦相似度计算,设定合理阈值,能够更快的得到闭环检测结果。最后通过在两个公开视觉SLAM闭环检测数据集,New College数据集和光照及角度变化更明显的City Centre数据集上进行实验,结果表明复杂环境下本文提出的方法比现有闭环检测方法,能够得到更高准确率和速率,更好满足了视觉SLAM系统对消除累计误差和实时性的要求。
For the camera perspective,lighting,climate,landform and other conditions of large changes or the existence of fast moving objects in the complex scene,the accuracy and robustness of simultaneous Locali-zation and mapping(SLAM)are low,a closed-loop detection as a solution to SLAM pose an important link of the drift,combined with this paper proposes a closed-loop detection method based on neural network.In this method,visual image data is obtained through sensors.Different from the traditional method of feature extraction,the improved triple constraint loss function is adopted in this paper to train Darknet to extract image features and construct corresponding feature vector matrix.Because Darknet borrowed from the idea of residuals network(RESNET),it has a deep network layer while maintaining a high accuracy rate,which greatly reduces the error of feature extraction.Through the self-encoder method to reduce the dimension of the data,through the cosine similarity calculation,set a reasonable threshold value,can get the closed-loop detection results faster.Last through the two open visual SLAM closed loop testing data set,the new college data collection and light and the angle change is more obvious city centre data set on the experiment,the results show that the proposed method under complicated environment than the existing closed loop detection method,can get higher accuracy and speed,and better meet the visual SLAM system to eliminate the accumulated error and real-time requirements.
作者
郭纪志
刘凤连
杨馨竹
汪日伟
GUO Jizhi;LIU Fenglian;YANG Xinzhu;WANG Riwei(Key Laboratory on Computer Vision and System,Ministry of Education of China,Key Laboratory on Intelligence Computing and Novel Software Technology of the City of Tianjin,Tianjin University of Thnology,Tianjin 300384,China;Computer Science and Engineering,Chongqing University of Technology,Chongqing 400054,China;Wenzhou University Oujiang College,Wenzhou,Zhejiang 325035,China)
出处
《光电子．激光》
CAS
CSCD
北大核心
2021年第6期628-636,共9页
Journal of Optoelectronics·Laser
基金
天津市教委科研重点项目(2017ZD13)资助项目。
关键词
视觉即时定位与地图构建
复杂场景
三重约束损失函数
闭环检测
自编码器
visual simultaneous localization and mapping
complex environment
convolutional neural network
loopback detection
autoencoders
