Study of Risk Avoidance Based on Classifying Motions of Humans near the Route for Driving Forklifts Autonomously in Warehouses

Abstract

In this research, we focus on the joint work with people of automated forklifts in the manned warehouse and aim the efficiency of the warehouse work by selecting the traveling operation of the forklift for the operation of the worker on the aisle and avoiding the danger Let's say. The motion of the person to set was assumed to be two motions of walking motion and motion to stay on shelf work, and the purpose was to separate it from moving images. The operation was reproduced in the warehouse, and a motion image was acquired by Kinect. People are detected from moving images, skeleton information is acquired by HOG feature and Open Pose , and binary classification is performed by SVM. We verified the accuracy of motion identification and compared it with the HOG feature and OpenPose on the accuracy change due to the change of the frame amount of image data and verified its effectiveness.