Joint Probabilistic Data Association fusion approach for pedestrian detection

Garcia, Fernando; de la Escalera, Arturo; Armingol, Jose Maria · 2013 · Crossref

DOI: 10.1109/ivs.2013.6629653

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of reliable pedestrian detection in Intelligent Transport Systems (ITS), where individual sensors often lack sufficient trustworthiness for road safety applications. The authors propose a novel decentralized fusion approach that combines data from a 2D laser scanner and a computer vision camera. The motivation stems from the complementary limitations of these sensors: computer vision provides rich but computationally expensive and unreliable unstructured data, while laser scanners offer reliable distance measurements but are limited to a single layer and struggle with classification. By integrating these sources using Joint Probabilistic Data Association (JPDA), a technique from Multiple Target Tracking theory, the system aims to enhance Advanced Driver Assistance Systems (ADAS) capabilities beyond what single-sensor or simpler fusion methods can achieve. The experimental setup utilizes a test platform equipped with a bumper-mounted laser scanner and a windshield-mounted camera, supplemented by GPS and inertial sensors for vehicle movement compensation. Low-level detection is performed independently for each sensor. The laser scanner identifies pedestrians based on a pattern of leg movement derived from consecutive polylines, validated over a sequence of detections. The camera system uses Histograms of Oriented Gradients (HOG) features, with the search region constrained by laser scanner data to reduce false positives. The fusion stage employs a Kalman Filter with a constant velocity model for state estimation. Crucially, the association of new detections to existing tracks is handled by JPDA, which calculates joint association probabilities for all possible measurement-to-track assignments, rather than using the simpler Global Nearest Neighbor (GNN) approach. Track management policies define consolidated tracks (confirmed by both sensors) and non-consolidated tracks, with specific logic to handle competing associations and prevent instability. The results demonstrate that the JPDA-based fusion significantly outperforms both single-sensor systems and fusion using GNN. Tested on over 10,000 frames across urban and interurban scenarios, the JPDA fusion achieved a positive detection rate of 82.29% with a misdetection rate of 1.11 per frame. In comparison, GNN fusion yielded 79.62% detection with 2.21 misdetections per frame, while standalone camera and laser systems had higher error rates. The paper highlights JPDA’s superior performance in challenging scenarios, including clustering errors where pedestrians merge into single obstacles, double detections caused by sensor misalignment or environmental noise, and occlusions where one pedestrian blocks another. In these cases, JPDA’s probabilistic weighting allowed for more accurate track maintenance and error correction compared to GNN, which often generated false tracks or failed to resolve ambiguities. The significance of this work lies in demonstrating that JPDA is a more robust and adaptable data association technique for pedestrian detection fusion than traditional methods like GNN. The decentralized scheme ensures system resilience, as detection can continue even if one sensor fails. The findings suggest that leveraging advanced tracking algorithms from Multiple Target Tracking theory can effectively overcome the inherent limitations of individual sensors, providing a more reliable solution for critical road safety applications. This approach offers a scalable framework for enhancing ADAS performance in complex, real-world driving environments.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• pedestrian behavior perception