Zheng Chaoda ; Yan Xu ; Gao Jiantao ; Zhao Weibing ; Zhang Wei ; Li Zhen ; Cui Shuguang

Abstract

Current 3D single object tracking approaches track the target based on afeature comparison between the target template and the search area. However,due to the common occlusion in LiDAR scans, it is non-trivial to conductaccurate feature comparisons on severe sparse and incomplete shapes. In thiswork, we exploit the ground truth bounding box given in the first frame as astrong cue to enhance the feature description of the target object, enabling amore accurate feature comparison in a simple yet effective way. In particular,we first propose the BoxCloud, an informative and robust representation, todepict an object using the point-to-box relation. We further design anefficient box-aware feature fusion module, which leverages the aforementionedBoxCloud for reliable feature matching and embedding. Integrating the proposedgeneral components into an existing model P2B, we construct a superiorbox-aware tracker (BAT). Experiments confirm that our proposed BAT outperformsthe previous state-of-the-art by a large margin on both KITTI and NuScenesbenchmarks, achieving a 15.2% improvement in terms of precision while running~20% faster.