基于Faster-RCNN的回环检测优化算法  

Optimized loop closure detection algorithm based on Faster-RCNN

作　　者：徐建鹏 卜凡亮 Xu Jianpeng;Bu Fanliang(School of Information Technology&Cyber Security,Chinese National Police University,Beijing 102628,China)

机构地区：[1]中国人民公安大学信息技术与网络安全学院

出　　处：《计算机应用研究》2019年第12期3628-3631,共4页Application Research of Computers

基　　金：北京自然科学基金项目(4184099);中国人民公安大学基础科研经费项目(2016JKF01316)

摘　　要：如何提高回环检测(loop closure detection)的准确率,是同时定位与地图构建系统(simultaneous localization and mapping,SLAM)中实现更高位姿恢复精度的关键问题之一。基于传统的词袋模型原理,构建了一个全新的算法框架。该算法使用预处理的Faster-RCNN对图像序列进行检测,利用所检测出的图像语义特征种类、像素位置及特征图等信息来构建具有标志性的二维语义特征向量图,并使用非线性的累积误差来计算二维语义特征向量图之间的相似度,且据此计算初始回环,经位姿验证后得到最终回环结果。通过与传统词袋模型算法的分析比较,实验结果验证了所提算法的有效性,实现了更高精度和效率的回环检测。How to improve the accuracy of loop detection is one of the key issues in the simultaneous positioning and map construction system to achieve higher pose recovery accuracy. Based on the traditional word bag model principle,this paper constructed a new algorithm framework. The algorithm used the pre-processed Faster-RCNN to detect the image sequence,and used the detected image semantic feature types,pixel positions and feature maps. To construct a two-dimensional semantic feature vector map with identification,and used nonlinear cumulative error to calculated the similarity between the two-dimensional semantic feature vector map,and calculated the initial loopback based on this,it obtained the final loopback result after pose verification. Compared with the traditional word bagging algorithm,the experimental results verify the effectiveness of the proposed algorithm and achieve more accurate and efficient loopback detection.

关 键 词：三维重建 同时定位与地图构建 回环检测 神经网络 词袋模型 优化算法 

分 类 号：D035.39[政治法律—政治学]