摘要
回环检测是视觉SLAM中的一个重要模块,成功检测出回环能够有效减少环境地图生成过程中的累积误差。针对传统方法主要利用人工设计特征,具有对光照变化非常敏感等问题,将深度学习算法运用于回环检测中,提出一种基于卷积神经网络的回环检测算法。利用预训练的卷积神经网络模型VGG16提取图像卷积特征,选取网络末端的池化层作为图像的全局特征表示,并通过感知哈希算法判断特征相似性,验证回环。从准确性和运算时间上在New college数据集上评估该算法的性能。实验结果表明,相对于传统算法,提出的算法有着更高的准确度和速率,准确度提高了27.9%,而特征提取时间减少了68.8%。证明了深度卷积神经网络对回环检测的有效性,能够更好地消除视觉SLAM系统的累积误差,同时具有更高的实时性。
Loop closure detection(LCD)is an important module in visual SLAM(simultaneous localization and mapping),which can effectively reduce the cumulative error in the environment map generation process. Since the traditional methods are very sensitive to illumination changes due to dependence of the artificial design features,the deep learning algorithm is applied to the LCD,and an LCD algorithm based on convolutional neural network(CNN) is proposed. The pre-trained CNN model VGG16 is used to extract image convolution features. The pooling layer at the end of the network is selected as the global feature representation of the image. The perceptual Hash algorithm is used to judge the feature similarity and verify the loop closure.The performance of the algorithm was evaluated on the data set New college in terms of accuracy and duration. The experimental results show that the proposed algorithm has higher accuracy and speed in comparison with the traditional algorithms. Its accuracy is increased by 27.9% and its feature extraction time is shortened by 68.8%. The experimental results have proved that the deep CNN is effective for LCD,and the LCD can better eliminate the cumulative error of the visual SLAM system and has higher real-time performance.
作者
李晓
马社祥
李啸
LI Xiao;MA Shexiang;LI Xiao(School of Electrical and Electronic Engineering,Tianjin University of Technology,Tianjin 300384,China;School of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China)
出处
《现代电子技术》
2022年第1期72-76,共5页
Modern Electronics Technique
基金
国家自然科学基金项目(61601326)
国家自然科学基金项目(61371108)。
作者简介
李晓,女,河南巩义人,硕士研究生,主要研究领域为计算机视觉、深度学习;马社祥,男,教授,主要研究领域为信号处理、图像处理等;李啸,男,博士研究生,主要研究领域为计算机视觉、深度学习。
