CHAMP: Crowdsourced, History-Based Advisory of Mapped Pedestrians for Safer Driver Assistance Systems

Greer, Ross; Rakla, Lulua; Desai, Samveed; Alofi, Afnan; Gopalkrishnan, Akshay; Trivedi, Mohan M. · 2023 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2301.05842

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 introduces CHAMP (Crowdsourced, History-Based Advisories of Mapped Pedestrians), a system designed to enhance driver assistance by predicting pedestrian presence in areas where visual detection fails, such as during nighttime or when pedestrians are occluded. The research is motivated by the limitations of current Automatic Emergency Braking (AEB) systems, which rely on real-time visual cues and often underperform in common scenarios like blind turns or low-light conditions. By leveraging aggregated historical data rather than single-instance detections, CHAMP aims to provide drivers with proactive advisories, thereby improving vigilance and mitigating collision risks in safety-critical situations. The methodology relies on two core principles: "Fleet" and "Repeat," which involve aggregating pedestrian detection data from repeated vehicle passes over specific geographic locations to identify statistical crossing patterns. The system comprises three stages: Pedestrian Location Association, which clusters GPS coordinates with detected pedestrians to build a map of hotspots; Nearest Pedestrian Hotspots Search, which uses a ball-tree algorithm to efficiently locate the nearest pedestrian cluster relative to the vehicle’s position and heading; and Advisory Issuance, which triggers warnings when the distance to a hotspot falls within a calculated stopping distance based on vehicle velocity and road friction. Data was collected in La Jolla, California, using a front-facing camera and GPS, resulting in 3 million frames annotated by human experts. Evaluation involved testing the system against ground truth in scenarios featuring occlusion, darkness, and blind turns, using precision and recall as primary metrics. The results demonstrate that CHAMP can achieve 100% precision and 75% recall on the experimental dataset when tuned with a 2-meter sampling distance. The system successfully issued advisories in test clips involving occluded pedestrians and dark lighting conditions, scenarios where traditional vision-based detectors typically fail. The authors note a trade-off between precision and recall, suggesting that further data collection and denoising techniques, such as trajectory filtering and time-of-day stratification, could optimize performance. Qualitative examples confirm that the system alerts drivers before they encounter pedestrians hidden from view, allowing for timely braking or caution. The significance of CHAMP lies in its potential to address a substantial portion of pedestrian incidents caused by visibility issues and improper crossing. The authors estimate that if integrated into navigation systems or autonomous vehicle APIs, CHAMP could mitigate up to 44,315 pedestrian incidents annually in the US. Beyond immediate safety benefits, the system offers applications for urban planning by mapping pedestrian foot traffic and supports privacy-preserving crowdsourcing through identity obscuration. This approach shifts the paradigm from reactive detection to proactive, history-based advisory, offering a scalable solution for enhancing road safety in complex urban environments.

Key finding

The CHAMP system achieved 100% precision and 75% recall in issuing pedestrian advisories on test data involving challenging scenarios like occlusions and dark lighting.

Methodology

lab_experiment

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. Discovered via author_sweep_intake on 2026-05-28.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

• pedestrian behavior perception