New Algorithm Uses Pedestrian Body Language to Guide Self-Driving Cars
Researchers at the University of Michigan develop a system that analyzes pedestrian pose and gait to better predict movements for autonomous vehicles.
Researchers at the University of Michigan develop an advanced algorithm that predicts pedestrian movements by analyzing body language rather than just tracking basic trajectories. While most autonomous vehicle vision systems simply identify a person and measure their speed, this new approach uses lidar and stereo cameras to evaluate specific details like pose and gait to determine a person's exact intentions.
This system detects subtle physical cues that indicate what a pedestrian plans to do next. For example, the technology notices if a person looks over their shoulder to cross the street, stares down at a phone, or puts their arms out to signal a stop, allowing the self-driving car to understand the context behind the movement.
The algorithm requires only a handful of frames, such as a single step and arm swing, to make highly accurate predictions that outperform simpler models. This quick processing capability proves essential for real-world driving where obstructions frequently hide pedestrians, making this complex body language analysis a critical step forward for autonomous vehicle safety.