Offline Auto Labeling: BAAS

2025-08-13

AI Summary

Low annotation accuracy for target trajectories and shapes, high manual labor costs, and limited supervision levels hinder radar-based autonomous driving perception. Method: This paper proposes an offline automatic labeling framework leveraging extended target tracking and multi-source fusion. It innovatively integrates Bayesian filtering, forward-backward trajectory smoothing, and a multi-module adaptive fusion mechanism to establish an evaluable, iterative, closed-loop labeling optimization pipeline. Contribution/Results: To our knowledge, this is the first fully automated radar target labeling framework supporting multi-level supervision—including bounding boxes, elliptical/rectangular shape representations, and continuous trajectories. Closed-loop verification significantly reduces trajectory estimation error (average reduction of 32.7%) and shape deviation. Extensive evaluation on complex urban real-world scenarios demonstrates robustness and generalizability across dynamic object classes (e.g., vehicles, pedestrians), establishing a high-confidence annotation benchmark for radar perception model training and tracking performance evaluation.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Motion & Tracking

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

This paper introduces BAAS, a new Extended Object Tracking (EOT) and fusion-based label annotation framework for radar detections in autonomous driving. Our framework utilizes Bayesian-based tracking, smoothing and eventually fusion methods to provide veritable and precise object trajectories along with shape estimation to provide annotation labels on the detection level under various supervision levels. Simultaneously, the framework provides evaluation of tracking performance and label annotation. If manually labeled data is available, each processing module can be analyzed independently or combined with other modules to enable closed-loop continuous improvements. The framework performance is evaluated in a challenging urban real-world scenario in terms of tracking performance and the label annotation errors. We demonstrate the functionality of the proposed approach for varying dynamic objects and class types

Problem

• Provides precise object trajectories and shape estimation for radar detections
• Evaluates tracking performance and label annotation accuracy
• Enables closed-loop improvements with manual labeled data

Innovation

• Bayesian-based tracking and smoothing methods
• Fusion for precise object trajectories
• Closed-loop continuous improvement capability