We address the problem of unsupervised semantic segmentation of outdoor LiDAR point clouds in diverse traffic scenarios. The key idea is to leverage the spatiotemporal nature of a dynamic point cloud sequence and introduce drastically stronger augmentation by establishing spatiotemporal correspondences across multiple frames. We dovetail clustering and pseudo-label learning in this work. Essentially, we alternate between clustering points into semantic groups and optimizing models using point-wise pseudo-spatiotemporal labels with a simple learning objective. Therefore, our method can learn discriminative features in an unsupervised learning fashion. We show promising segmentation performance on Semantic-KITTI,SemanticPOSS, and FLORIDA benchmark datasets covering scenarios in autonomous vehicle and intersection infrastructure, which is competitive when compared against many existing fully supervised learning methods. This general framework can lead to a unified representation learning approach for LiDAR point clouds incorporating domain knowledge.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    A Spatiotemporal Correspondence Approach to Unsupervised LiDAR Segmentation with Traffic Applications


    Contributors:
    Li, Xiao (author) / He, Pan (author) / Wu, Aotian (author) / Ranka, Sanjay (author) / Rangarajan, Anand (author)


    Publication date :

    2023-09-24


    Size :

    3830230 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Unsupervised multiphase segmentation: A recursive approach

    Ni, K. / Hong, B. W. / Soatto, S. et al. | British Library Online Contents | 2009



    Spatiotemporal Object Database Approach to Dynamic Segmentation

    Huang, Bo / Yao, Li | Transportation Research Record | 2003


    Unsupervised Alignment of Traffic Elements for Domain Adaptive Semantic Segmentation

    Gao, Yuan / Li, Yaochen / Zhang, Tengwen et al. | IEEE | 2024