Authors: Maria Lyssenko, Piyush Pimplikar, Maarten Bieshaar, Farzad Nozarian, Rudolph Triebel

Published on: February 05, 2024

Impact Score: 8.12

Arxiv code: Arxiv:2402.02986

Summary

• What is new: Introduction of a novel safety-aware loss variation that incorporates per-pedestrian criticality scores in the training process of object detectors in automated driving systems.
• Why this is important: Existing object detection models in automated driving systems do not differentiate between mistakes based on their safety criticality, treating all misdetections equally.
• What the research proposes: A safety-aware training strategy that uses a new loss function considering the criticality of pedestrians, thereby prioritizing the detection of safety-critical cases.
• Results: Enhanced detection of critical pedestrians without compromising overall performance, as demonstrated by evaluations on the nuScenes dataset with RetinaNet and FCOS models.

Technical Details

Technological frameworks used: RetinaNet, FCOS

Models used: Deep neural networks (DNN)

Data used: nuScenes dataset

Potential Impact

Automated driving systems, autonomous vehicle manufacturers, road safety equipment producers, and AI-based object detection technology companies

Want to implement this idea in a business?

We have generated a startup concept here: SafeguardAI.