面向室内动态场景的多传感视觉SLAM方法  

Multi-sensor Visual SLAM Method for Indoor Dynamic Scenes

作　　者：张建华[1] 张天晶 赵岩[1] 张霖 周浩 ZHANG Jianhua;ZHANG Tianjing;ZHAO Yan;ZHANG Lin;ZHOU Hao(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300132,China)

机构地区：[1]河北工业大学机械工程学院,天津300132

出　　处：《信息与控制》2022年第6期641-650,661,共11页Information and Control

基　　金：国家重点研发计划(2018YFB1306500);河北省自然科学基金(F2017202062,F2020202053);天津市自然科学基金(19JCJQJC61600)。

摘　　要：针对传统同时定位与地图构建(simultaneous localization and mapping,SLAM)框架面临动态场景时产生明显定位误差,建立的场景稠密地图会包含动态对象及其运动叠影,从而导致定位与建图鲁棒性不足的问题,面向以人类为主要动态对象的室内动态场景,从“温度”的角度出发,提出基于热像仪与深度相机结合的多传感SLAM协同方案,解决室内动态场景中的定位与建图难题。首先,建立一套针对热像仪与深度相机的联合标定策略,重新设计标定板与标定方案,完成相机的内参标定、外参标定与图像配准,得到一一对应的RGB、深度、热(RDH)三模图像;其次,由热图像得到人体掩模图像,进而在ORB-SLAM2系统框架下构建静态特征提取与数据关联策略,实现基于三模图像的视觉里程计;然后,基于人体掩模图像更新深度图像,滤除人体区域,进而完成基于三模图像的静态环境稠密地图构建;最后,在室内动态场景下进行实验验证,结果表明所提出算法在室内动态场景下可有效剔除动态对象的干扰特征,相对传统SLAM算法具有明显优势。Owing to the localization error associated with traditional SLAM frameworks used in dynamic scenes,the established scene dense map contains dynamic objects with overlapping motions,leading to insufficient robustness of localization and mapping.For indoor dynamic scenes with humans as the main dynamic object,and from the perspective of“temperature”,we propose a collaborative scheme of multi-sensor SLAM using a combination of the thermal imaging system and depth camera to tackle the issue of positioning and mapping in indoor dynamic scenes.First,we establish a set of joint calibration strategies for the thermal imager and depth camera,redesign the calibration plate and scheme to complete the internal parameter calibration,external parameter calibration of the camera and image registration,and obtain the one-to-one corresponding three-mode images of RGB,depth,and heat(RDH).Second,we use the thermal image to obtain the mask image of the human body.Then,we construct the static feature extraction and data association strategy under the framework of the ORB-SLAM2 system to realize the visual odometry using a three-mode image.The body region is filtered based on the updated depth image,and the static environment density map based on the three-mode image is constructed.Finally,the experiments are performed in a dynamic indoor scene,and the results indicate that the proposed algorithm effectively eliminates the interference features of dynamic objects in a dynamic indoor scene and has obvious advantages over traditional SLAM algorithms.

关 键 词：动态场景 视觉SLAM 热像仪 RDH三模图像 多传感协同 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]