Content area
Roadway intersections are among the most hazardous locations for pedestrians due to the complex interactions between vehicles and pedestrians, especially during right-turn maneuvers. Human drivers and traditional vehicles are both constrained by visibility and rely on intuition, eye contact, and quick judgment to respond to sudden pedestrian movement. Autonomous Vehicles (AVs), however, are dependent exclusively on sensor inputs and algorithmic processing, and thus are more prone to detection problems when line-of-sight is obstructed.
This thesis examines the challenge that Autonomous Vehicles (AVs) face in detecting pedestrians at signalized intersections with limited sight distances and explores how Vehicle-to-Everything (V2X) communication can mitigate detection blind spots. Using a MATLAB simulation, this work simulated a real-world intersection in Jersey City, NJ, incorporating autonomous vehicle (AV) dynamics, pedestrians, and environmental occlusions.
Findings revealed that under existing sight distance conditions (13 ft), AVs detected pedestrians at only 14% of points, with a collision probability of 1.0 for distances of less than 14.8 ft. Using the regulatory corner sight distance of 100 feet, detection was raised to 58%, which significantly reduced collision dangers. Additionally, V2V communication enabled occlusion-free, real-time pedestrian warning among vehicles, while V2I systems enhanced early warning through roadside infrastructure. Furthermore, V2P communication, facilitated by Class 1 Bluetooth, enabled occlusion-free, timely pedestrian detection.
This research offers policy-critical results for infrastructure policy, AV design, and urban planning, recommending V2X integration and intersection design modifications as imperative steps toward safer pedestrian-AV interaction.
