A Stereo Camera Based Static and Moving Obstacles Detection on Autonomous Visual Navigation of Indoor Transportation Vehicle

Abstract

This paper proposes an indoor moving obstacle avoidance system with stereo camera for an autonomous transportation vehicle. Using three-dimension (3D) data, the obstacle avoidance system first separates the floor and non-floor data. Then, using density-based spatial clustering of applications with noise, the system classifies obstacles as static or moving based on optical flows from the data's center of gravity. Next, using Enhanced Vector Field Histogram method on depth data, the system locates a moving space for the vehicle and estimates the velocity and angular velocity of any moving obstacles using Kalman filter. In addition, the system selects a space for passing with Enhanced Vector Field Histogram method only for static obstacles. The system predicts whether avoidance is possible based on estimated velocity of obstacle. If obstacles cannot be avoided, another path is selected. The proposed system is validated in experiments using a moving robot as the test vehicle in traffic situations where collision with other moving objects (i.e., obstacles) is predicted. The results show that the robot effectively avoids moving obstacles by turning or changing velocity.