In autonomous vehicle navigation, the effectiveness and robustness of object detection models are directly influenced by the quality, quantity, and diversity of data. Detecting Vul-nerable Road Users (VRU) poses significant challenges due to their constant motion and dynamic behaviors. Utilizing real-time datasets like KITTI for training VRU detection models may not be optimal, as they lack coverage of adversarial situations and undervalue objects like pedestrians. Relying solely on such datasets for model training can lead to catastrophic real-world results. To address these issues, we present PointGAN, a framework supporting 3D object detection models that leverage conditional Generative Adversarial Networks (cGAN) to enhance dataset diversity specifically for the pedestrian class. PointGAN employs generative neural networks trained through multiple iterations to generate realistic point cloud data, guided by feedback from the discriminator until it closely mirrors real-world data. This strategy intends to significantly improve the overall performance of 3D object detection models by skillfully detecting pedestrians in challenging and diverse scenarios. The generative model achieves a Minimum Matching Distance - Earth Mover's Distance (MMD-EMD) score of 0.025, outperforming the existing state-of-the-art models trained under different categories.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    PointGAN: A Catalyst for Enhanced Vulnerable Road User Detection in Autonomous Navigation




    Publication date :

    2024-06-24


    Size :

    659153 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English




    VULNERABLE ROAD USER IDENTIFICATION SYSTEM

    ALTEKAR NIRAJ / RYAN MICHAEL / RAO JAYANTHI | European Patent Office | 2025

    Free access

    ENHANCED VULNERABLE ROAD USER (VRU) PREDICTION THROUGH CLOUD-BASED PROCESSING

    VASSILOVSKI DAN | European Patent Office | 2024

    Free access

    ENHANCED VULNERABLE ROAD USER (VRU) PREDICTION THROUGH CLOUD-BASED PROCESSING

    VASSILOVSKI DAN | European Patent Office | 2024

    Free access