Research on Geometry-Based Algorithm to Avoid Collisions with Pedestrians for Autonomous Vehicles

Autonomous cars have become increasingly popular in recent years. There are many issues associated with autonomous vehicles, among which collision avoidance with pedestrians is a crucial concern. Regarding humans, ensuring safety requires both processing speed and accuracy. This paper introduces an algorithm that uses only a camera to detect lane lines and avoid pedestrians. The main contribution of this paper is based on geometry, which determines whether a pedestrian is crossing the street or not and calculates the steering wheel to avoid pedestrians based on math. The algorithm is divided into four stages. In the first stage, YOLOV7 was utilized to detect pedestrians. In the second stage, Sort was used to track people. In the third stage, PIDNet is deployed to determine the lane for autonomous vehicles and to ascertain whether a pedestrian is crossing the street, leveraging geometric analysis. In the last stage of the method, the steering to avoid a pedestrian collision was calculated. Finally, the algorithm conducted experiments on golf cars in an outdoor environment, achieving the results with high accuracy. The demo video is attached: https://drive.google.com/drive/folders/1fyf8_3ig1RAhW1ItK5icGdBCph_GT5OI?usp=sharing .