Advanced Multi-Modal Sensor Fusion System for Detecting Falling Humans: Quantitative Evaluation for Enhanced Vehicle Safety

Collisions with fallen pedestrians pose a lethal challenge to current advanced driver-assistance systems. This paper introduces and quantitatively validates the Advanced Falling Object Detection System (AFODS), a novel safety framework designed to mitigate this risk. AFODS architecturally integrates long-wave infrared, near-infrared stereo and ultrasonic sensors, processed through a novel artificial intelligence pipeline that combines YOLOv7-Tiny for object detection with a recurrent neural network for proactive threat assessment, thereby enabling the system to predict falls before they are complete. In a rigorous controlled study using simulated adverse conditions, AFODS achieved a 98.2% detection rate at night, a condition where standard systems fail. This paper details the system’s ISO 26262-aligned architecture and validation results, proposing a framework for a new benchmark in active vehicle safety, demonstrated under controlled test conditions.