Now, most existing dynamic RGB-D SLAM methods are based on deep learning or mathematical models. Abundant training sample data is necessary for deep learning, and the selection diversity of semantic samples and camera motion modes are closely related to the robust detection of moving targets. Furthermore, the mathematical models are implemented at the feature-level of segmentation, which is likely to cause sub or over-segmentation of dynamic features. To address this problem, different from most feature-level dynamic segmentation based on mathematical models, a non-prior semantic dynamic segmentation based on a particle filter is proposed in this paper, which aims to attain the motion object segmentation. Firstly, GMS and optical flow are used to calculate an inter-frame difference image, which is considered an observation measurement of posterior estimation. Then, a motion equation of a particle filter is established using Gaussian distribution. Finally, our proposed segmentation method is integrated into the front end of visual SLAM and establishes a new dynamic SLAM, PFD-SLAM. Extensive experiments on the public TUM datasets and real dynamic scenes are conducted to verify location accuracy and practical performances of PFD-SLAM. Furthermore, we also compare experimental results with several state-of-the-art dynamic SLAM methods in terms of two evaluation indexes, RPE and ATE. Still, we provide visual comparisons between the camera estimation trajectories and ground truth. The comprehensive verification and testing experiments demonstrate that our PFD-SLAM can achieve better dynamic segmentation results and robust performances.

Most existing visual simultaneous localization and mapping (SLAM) algorithms rely heavily on the static world assumption. Combined with deep learning, semantic SLAM has become a popular solution for dynamic scenes. However, most semantic SLAM methods show poor real-time performance when dealing with dynamic scenes. To handle this problem, a real-time semantic SLAM method is proposed in this paper, combining knowledge distillation and dynamic probability propagation strategy. First, to improve the execution speed, a multi-level knowledge distillation method is adopted to obtain a lightweight segmentation model, which is more suitable for continuous frames to create an independent semantic segmentation thread. This segmentation thread only accepts keyframes as input so that the system can avoid time delay caused by processing each frame. Second, a static semantic keyframe selection strategy is proposed based on the segmentation results. In this way, those keyframes containing more static information will be selected to reduce the participation of dynamic objects. By combining segmentation results and data matching algorithm, our system can realize the update and propagation of dynamic probability, reducing the influence of dynamic points in the pose optimization process. Validation results based on the KITTI and TUM datasets show that our method can effectively deal with dynamic feature points and improve running speed simultaneously.