Cars use cameras, radar, and Light Detection and Ranging (LiDAR) to locate traffic, but cameras struggle in darkness and LiDAR can falter in bad weather. Microphones may still capture sounds that reveal nearby vehicles.
The team mounted microphones around a car and recorded real drives, using LiDAR to mark vehicle positions. AcousticBEV turns the sound into vehicle-center estimates on a bird’s-eye view, a map-like view from above, while accounting for the car’s speed.
It outperformed the tested conventional and learning-based audio methods, with F1 scores, which balance correct detections against false alarms and misses, of 47.7% in easier scenes and 45.1% in harder ones. The results show microphones can add direction and distance information to vehicle sensing.
