Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

186 Views

Abstract for presentation (Poster or Podium) with a Paper in the Conference Proceedings

AI in Transportation

Multi-camera Vehicle Re-identification for Privacy-aware Network-level Travel Time Estimation

Monday, June 17, 2024

1:30 PM - 2:45 PM ET

Room: Galleria

Primary Author(s)

CL

Chenxi Liu

Ph.D. Candidate
University of Washington
Seattle, WA, United States

Presenter(s)

CL

Chenxi Liu

Ph.D. Candidate
University of Washington
Seattle, WA, United States

Co-Author(s)

Zening Wang, n/a

Undergraduate Student
University of California - San Diego
San Diego, California, United States
Hao Yang, PhD Candidate

PhD candidate
University of Washington
Seattle, WA, United States
Torgeir Vaa, MSc

Senior Principal Engineer
Norwegian Public Roads Administration
Trondheim, Sor-Trondelag, Norway
Yinhai Wang

Ph.D., P.E.
Univ. of Washington
Seattle, WA, United States

Submitter(s)

CL

Chenxi Liu

Ph.D. Candidate
University of Washington
Seattle, WA, United States

Abstract: Traffic surveillance cameras, favored for their cost-effectiveness and data richness, have rapidly permeated our urban landscapes. Yet, the true potential of these systems remains elusive, largely because cameras function in silos, constrained by their individual viewpoints. Compounding this, rising privacy concerns challenge the public's acceptance of such surveillance measures. Responding to these challenges, we present an edge-based Multi-camera Vehicle Re-identification system, a groundbreaking framework devised for network-level travel time estimation through vehicle re-identification across multiple cameras. Leveraging the fast-reid algorithm, our team adeptly tailored it for seamless execution on portable edge AI devices without compromising on accuracy or reliability. Rigorous evaluations on diverse vehicle datasets manifested a remarkable recognition success rate, peaking at 98% post-training. When transitioned to edge devices, this algorithm maintained its stellar performance, exemplified through consistent ROC curves and the ability to compute and assess safe vehicle travel times. The envisioned scale-up of this technology suggests the feasibility of expansive multi-camera networks with the capability to monitor vehicle behaviors autonomously. Only anomalous driving patterns will prompt real-time alerts, thereby streamlining manpower demands. This precision promises swift interventions - from prompt investigations to potential search and rescue missions, thereby amplifying safety and operational efficiency in our transport systems.

Learning Objectives:

Attendees can expect to learn the following from this session:

Understand the challenges and limitations posed by isolated operation of traffic surveillance cameras and the emerging privacy concerns affecting their widespread acceptance.
Understand the capabilities and features of the edge-based Multi-camera Vehicle Re-identification system, focusing on its ability to estimate network-level travel time through vehicle re-identification across various cameras.
Understand the implications of its deployment regarding manpower efficiency, safety enhancement, and operational efficiency in transport systems, particularly in the context of real-time alerts and rapid interventions.