IMAGE RECOGNITION AS A SUITABLE HUMANOID ROBOT IMITATION

Abstract

We describe in this work a system that uses image recognition to simulate human movements on a
humanoid robot. Three primary procedures need be taken into consideration in order to accomplish this
imitation. These include data collection, data manipulation, and ankle angle adjustment on the robot's
supporting foot. In the first step of the human imitation data collecting procedure, 13 color markings are
pasted at various points on a human body. These color markings must be identified using a Logitech
camera C905 in order to get the motion imitation. The relative locations of the marks for each motion
are then recorded and computed into the motion database. To ensure that the humanoid robot's zero
moment point is inside the stable zone, these data are altered during the data modification stage with the
use of computer simulation. The last step, known as the ankle angle adjustment stage, involves adjusting
the ankle angles. As a result, the humanoid robot can accurately and steadily mimic almost every human
movement in real time.