基于几何约束和目标检测的室内动态SLAM

针对即时定位与地图构建(SLAM)在室内动态环境下定位精度低和地图效果差的问题，提出一种基于几何约束和目标检测的室内动态SLAM方法. 使用目标检测网络获取语义信息，提出运动物体漏检的方法；根据先验知识，提出准确识别动态区域的信息判定方法；结合几何约束和深度学习方法剔除动态点，利用静态点估计相机位姿；根据存储信息构建可闭环的静态地图. 在TUM数据集上进行实验，定位精度比ORB-SLAM2提高97.5%，相较于其他动态SLAM可取得更好的性能. 在室内真实环境进行实验，构建的静态地图更准确，有效提高了室内动态SLAM的定位精度和地图效果. 

Indoor dynamic SLAM based on geometric constraints and target detection

Aiming at the problem of low localization accuracy and poor map effect of visual simultaneous localization and mapping (SLAM) in indoor dynamic environment, a indoor dynamic SLAM method is proposed based on geometric constraints and target detection. The target detection network is used to obtain semantic information and a method for missing detection of moving objects is proposed. Based on prior knowledge, an information determination method is proposed to accurately identify dynamic regions. Dynamic points are eliminated based on geometric constraints and deep learning. Static points are used to estimate camera pose. A closed-loop static map is builded based on the stored information. The experiment on TUM dataset shows that the localization accuracy is 97.5% higher than that of ORB-SLAM2 and the performance is better than other dynamic SLAM. The experiment in the indoor real environment shows that the static map is more accurate. The localization accuracy and the map effect of indoor dynamic SLAM are improved effectively.