[ Robot vision] Stereo Based Human Detection Tracking

Realtime Stereo-based Human Detection and Tracking for Robot-Vision.

Introduction

Given RGB-D sensory inputs, we design 4 directional 2D elliptical filters (4D2DEFs) and convolve them on the disparity map (in x-z space) to localize multiple humans. Resulting candidate blobs are further verified by either detecting the face or matching head-shoulder shapes. To preserve a person’s ID, we make use of particle filtering for robust tracking. Note that this algorithm is dedicated to applying mobile robots so that we do not use any background modeling technique, and works in real-time.

Keywords:

Robot Vision, Human Detection/Tracking

To get further details, check out our paper:
Pose Robust Human Detection in Depth Image Using Four Directional 2D Elliptical Filters, IEEE ISM, 2009