Paper Acceptance: Perception Metrics for Intelligent Vehicles at IEEE ITSC 2025

Galia Fabiola Cornejo Urquieta, Javier Ibanez-Guzman, Federico Camarda, Julien Moreau, Yves Grandvalet

Abstract

Standard computer vision metrics are not sufficient for safety-critical applications like Automated Emergency Braking (AEB). This paper benchmarks uncertainty estimation methods (Deep Ensembles, Monte Carlo Dropout) under diverse conditions and highlights the need for a unified framework that blends vision-oriented and automotive-centric metrics.

Date
Jul 1, 2025 —
Event
IEEE International Conference on Intelligent Transportation Systems (ITSC)
This paper has been accepted for presentation at the IEEE International Conference on Intelligent Transportation Systems (ITSC) 2025.

Overview

We show that standard computer vision metrics are insufficient for safety-critical applications like Automated Emergency Braking (AEB).

Our research highlights:

  • Use Case: A specific focus on AEB and safety-critical scenarios.
  • The Gap: How two detections can have the same IoU but drastically different distance estimation errors.
  • Benchmarking: An evaluation of uncertainty estimation methods (Deep Ensembles, Monte Carlo Dropout) under diverse conditions.
  • Solution: The need for a unified framework blending vision-oriented and automotive-centric metrics.

This work is a collaboration between Renault Group (Ampere Software Technology) and Heudiasyc (UTC - CNRS).

AEB perception metrics uncertainty estimation autonomous driving safety-critical systems
Federico Camarda
Federico Camarda
AD/ADAS Sofware Engineer | PhD in Automation and Robotics

My work and research aim is to make vehicles smarter and safer.