Edge AI for Automotive Vulnerable Road User Safety: Deployable Detection via Knowledge Distillation

TL;DR

This paper introduces a knowledge distillation framework that enables deploying compact, accurate object detection models for VRU safety on edge hardware, overcoming quantization challenges and maintaining high precision.

Contribution

It presents a novel KD method that trains small YOLOv8 models to mimic larger ones, preserving accuracy under INT8 quantization for edge deployment.

Findings

KD student retains 5.6% less mAP than teacher under INT8, outperforming direct training.

KD transfers calibration, leading to 14.5% higher precision at similar recall.

KD student surpasses teacher in FP32 precision despite being 3.9x smaller.

Abstract

Deploying accurate object detection for Vulnerable Road User (VRU) safety on edge hardware requires balancing model capacity against computational constraints. Large models achieve high accuracy but fail under INT8 quantization required for edge deployment, while small models sacrifice detection performance. This paper presents a knowledge distillation (KD) framework that trains a compact YOLOv8-S student (11.2M parameters) to mimic a YOLOv8-L teacher (43.7M parameters), achieving 3.9x compression while preserving quantization robustness. We evaluate on full-scale BDD100K (70K training images) with Post-Training Quantization to INT8. The teacher suffers catastrophic degradation under INT8 (-23% mAP), while the KD student retains accuracy (-5.6% mAP). Analysis reveals that KD transfers precision calibration rather than raw detection capacity: the KD student achieves 0.748 precision versus 0.653 for direct training at INT8, a 14.5% gain at equivalent recall, reducing false alarms by 44% versus the collapsed teacher. At INT8, the KD student exceeds the teacher's FP32 precision (0.748 vs. 0.718) in a model 3.9x smaller. These findings establish knowledge distillation as a requirement for deploying accurate, safety-critical VRU detection on edge hardware.