The development and testing of automotive perception systems require a large amount of labeled data. The ground truth of the measurement scene is usually generated by manual video annotation. However, besides requiring a lot of manual effort, it can be time-consuming. Moreover, since the annotation boxes must be transformed into the vehicle's BEV space, the result is often not precise enough due to the camera's poor depth estimation. The existing ground truth generation systems rely on extremely expensive sensor clusters, including LiDARs and GNSS RTK. In this paper, we present a novel semi-automatic ground truth generation tool that exclusively depends on a mono-vision camera and an automotive radar, eliminating the need for these costly sensors. The proposed system comprises an automatic pre-generation algorithm and an interactive GUI annotation tool. We present a novel pre-generation algorithm relying on offline tracking algorithms and fusing radar and camera data that exploits the offline nature of the problem and enhances the preciseness of depth estimation. The proposed system is evaluated on a highway section by comparing the automatically generated objects to the manually created ground truth. According to the results, the automatic pre-generation can significantly decrease the effort of manual annotation, which is also supported by a user-friendly GUI. Along with the theoretical background, the tool and the dataset used for this paper are provided on GitHub.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Semi-Automatic BEV Ground Truth Generation for Automotive Perception Systems


    Contributors:

    Published in:

    Publication date :

    2025-02-01


    Size :

    3316713 byte




    Type of media :

    Article (Journal)


    Type of material :

    Electronic Resource


    Language :

    English



    Semi-Automatic Ground Truth Trajectory Estimation and Smoothing using Roadside Cameras

    Fleck, Tobias / Zipfl, Maximilian / Zollner, J. Marius | IEEE | 2023


    GROUND TRUTH DATA GENERATION FOR DEEP NEURAL NETWORK PERCEPTION IN AUTONOMOUS DRIVING APPLICATIONS

    TILMAN WEKEL / JOACHIM PEHSERL / JACOB MEYER et al. | European Patent Office | 2022

    Free access

    Evaluation of two-view geometry methods with automatic ground-truth generation

    Lakemond, R. / Fookes, C. / Sridharan, S. | British Library Online Contents | 2013


    Deep learning based multi-modal perception and semi-automatic labelling algorithms for automotive sensor data

    Haase-Schütz, Christian / kassel university press | TIBKAT | 2024

    Free access

    Inducing and obtaining cognitive load ground truth data in automotive scenarios

    Sultana, Alina E. / Nicolae, Irina E. / Fulop, Szabolcs et al. | SPIE | 2023