Loop closure can effectively eliminate the accumulated error and plays an important role in Simultaneous Localization and Mapping (SLAM). There remains challenges in loop detection and loop correction due to the large viewpoints difference and the environment appearance changes. In this letter, we propose a novel loop closure method based on object modeling and semantic graph matching. We use voxels and cuboids to model the object-level features in the environment, and further represent the environment as a semantic graph with topological information. On this basis, an efficient graph matching method based on edit distance is proposed for robust place recognition. Finally, the loop correction is carried out by object alignment between semantic maps. Experimental results demonstrate that the proposed method can realize omni-directional loop closure without viewpoint constraint, and is robust to environmental appearance changes. Compared with the existing appearance-based methods, the performance has been greatly improved.

Accurate trajectory prediction of vehicles is essential for reliable autonomous driving. To maintain consistent performance as a vehicle driving around different cities, it is crucial to adapt to changing traffic circumstances and achieve lifelong trajectory prediction model. To realize it, catastrophic forgetting is a main problem to be addressed. In this paper, a divergence measurement method based on conditional Kullback-Leibler divergence is proposed first to evaluate spatiotemporal dependency difference among varied driving circumstances. Then based on generative replay, a novel lifelong vehicle trajectory prediction framework is developed. The framework consists of a conditional generation model and a vehicle trajectory prediction model. The conditional generation model is a generative adversarial network conditioned on position configuration of vehicles. After learning and merging trajectory distribution of vehicles across different cities, the generation model replays trajectories with prior samplings as inputs, which alleviates catastrophic forgetting. The vehicle trajectory prediction model is trained by the replayed trajectories and achieves consistent prediction performance on visited cities. A lifelong experiment setup is established on four open datasets including five tasks. Spatiotemporal dependency divergence is calculated for different tasks. Even though these divergence, the proposed framework exhibits lifelong learning ability and achieves consistent performance on all tasks.

Objects in environments are more stable than feature points, which can benefit for building a wide range of common view relationship and effectively reduce the cumulative error of visual localization. In this paper, we propose Contour-SLAM, the first RGB-D object-level SLAM method that utilizes the projection constraints between the dense object model and its instance masks for state estimation. Firstly, considering that the unstructured object models are difficult to describe mathematically and their re-projection errors are difficult to quantify, we use voxels and cuboids to model objects and define the re-projection error between the 3D object model and its 2D instance segmentation mask based on the outer contour distance. Secondly, we construct a covisibility graph between keyframes based on the shared observations of same feature points and objects, which is further used to maintain a local map including keyframes, map points and object models in real-time. Finally, a multi-view bundle adjustment formulation is proposed to jointly optimize the three components of the local map. Experimental results on public dataset as well as our own collected sequences demonstrate that object modeling and camera pose estimation can benefit each other. The proposed method demonstrates competitive performance compared with the state-of-the-art SLAM methods.