Abstract:

This paper proposes a monocular vision solution integrated with industrial robots to determine the spatial position and pose of objects within unstructured working environments. A significant challenge associated with 2D cameras is the perspective distortion error that arises when the optical axis is not perpendicular to the object's surface. To address this issue, this study develops an algorithm based on the Homography matrix to accurately estimate the Tilt Angle between the camera axis and the surface normal, while simultaneously calculating the distance and Lateral Offset using the Pinhole Camera Model. The system is implemented on the Robot Operating System (ROS) platform utilizing a distributed Publisher/Subscriber architecture, which enables the decoupling of image processing and kinematic control workflows. Experimental results on a Niryo Ned 2 robot with basic geometric objects demonstrate that the system operates stably in real-time (30Hz), achieves a distance estimation error of less than 2%, and accurately determines the tilt angle to facilitate robot approach pose compensation. This solution offers a cost-effective and highly integrable approach for automation applications in the mining industry.