StreamMOS: Streaming Moving Object Segmentation With Multi-View Perception and Dual-Span Memory

Abstract

Moving object segmentation based on LiDAR is a crucial and challenging task for autonomous driving and mobile robotics. Most approaches explore spatio-temporal information from LiDAR sequences to predict moving objects in the current frame. However, they often focus on transferring temporal cues in a single inference and regard every prediction as independent of others. This may lead to inconsistent segmentation results for the same object across different frames. To solve this issue, we propose a streaming network with a memory mechanism, called StreamMOS, to build the association of features and predictions among multiple inferences. Specifically, we utilize a short-term memory to convey historical features, which can be regarded as spatial priors of moving objects and are used to enhance current inference by temporal fusion. Meanwhile, we build a long-term memory to store previous predictions and exploit them to refine current forecasts at the voxel and instance levels through voting. Besides, we apply multi-view encoder with cascaded projection and asymmetric convolution to extract motion feature of objects in different representations. Extensive experiments validate that our algorithm gets competitive performance on SemanticKITTI and Sipailou Campus datasets.