Edge Detection and ORB Matching for Accurate Pedestrian Distance Estimation Using Stereovision

Abstract

Pedestrian detection is a crucial component of Advanced Driver Assistance Systems (ADAS), enhancing safety and collision prevention. For effective safety solutions, accurate distance estimation must complement detection efforts. However, research focusing specifically on distance estimation for pedestrians using only stereovision is limited, with most studies relying on computationally intensive dense depth maps. This paper presents a novel method for pedestrian distance estimation using stereovision, designed to minimize computational load. The technique integrates edge detection based on Sobel magnitude filtering with ORB (Oriented FAST and Rotated BRIEF) matching to compute object-level disparity. Validation of the proposed method in real-time on a Raspberry Pi 4 demonstrates its feasibility on resource-constrained device. Experimental results indicate that the approach achieves reliable and efficient distance estimation for pedestrians, providing a significant advancement in stereovision-based safety systems.