多模态嵌入与轨迹修正的三维多目标跟踪. (Chinese)

3D multi-object tracking with multi-modal embedding and trajectory correction. (English)

The multi-modal features of point clouds and images have strong complementary advantages and can effectively improve the performance of three-dimensional multi-object tracking. However, object tracking still faces many challenges due to the complexity of the tracking scene and the uncertainty of the object state. Based on this, this paper proposed a three-dimensional multi-object tracking algorithm with multi-modal embedding and trajectory correction. Firstly, it constructed a multi- modal embedding learning network to learn more discriminative embedding representations through multi-scale semantic feature learning and multi-modal re-fusion modules. Secondly, it designed a multi-feature comprehensive correlation module to jointly track embedding and geometric information, while correcting angle prediction errors to achieve more accurate data correlation. Finally, it proposed a dual-stream trajectory correction and management algorithm to correct erroneous disappearing trajectories to improve trajectory accuracy. The proposed method was evaluated on the KITTI data set and compared with other advanced methods. The HOTA index of the proposed method reached 77.72%, and the MOTA index reached 88.24%, showing the best tracking performance overall. Experiments prove that the proposed method effectively improves tracking accuracy, reduces the occurrence of tracking errors, and has good tracking performance. [ABSTRACT FROM AUTHOR]

点云和图像的多模态特征具有很强的优势互补性, 能够有效提升三维多目标跟踪的性能。然而, 由于 跟踪场景的复杂性和目标状态的不确定性, 使得目标跟踪仍面临许多挑战。基于此, 提出多模态嵌入与轨迹修 正的三维多目标跟踪算法。首先构建多模态嵌入学习网络, 通过多尺度语义特征学习与多模态再融合模块, 学 习更具判别性的嵌入表示; 其次, 提出多特征综合关联模块, 联合跟踪嵌入和几何信息, 同时修正角度预测错误, 实现更精确的数据关联; 最后, 提出双流轨迹修正与管理算法, 修正错误消失轨迹, 以提升轨迹的准确性。在 KITTI 数据集上对提出的方法进行评估并与其他先进方法进行比较, 该方法的HOTA指标达到了77.72%, MO-TA 指标达到了88.24%, 整体体现出较好的跟踪性能。实验证明该方法有效地提升了跟踪精度, 并减少了跟踪 错误的发生, 具有良好的跟踪性能。 [ABSTRACT FROM AUTHOR]

Copyright of Application Research of Computers / Jisuanji Yingyong Yanjiu is the property of Application Research of Computers Edition and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)