A Low-Cost Modular Vision System for Flexible and Intelligent Industrial Robots towards Autonomous Manufacturing

Yang, E. (Speaker)

Activity: Talk or presentation typesInvited talk

Description

With Industry 4.0 being currently widely acknowledged as a key driver of industrial advancement, a need for a strong technological shift has become apparent within manufacturing industry to move towards more intelligence and autonomy. Machine vision plays a significant role in enabling traditionally industrial robots flexible and intelligent to meet their needs for applications in autonomous manufacturing. In this talk, a low-cost modular vision system for object detection and motion tracking in a challenging environment with non-uniform changing conditions is presented. A practical case study is conducted on industrial robots (KUKA KR90 R3100 model) to perform real-time path planning and optimize its route while avoiding a moving obstacle in the workspace. The developed vision system is able to detect and track the dynamic obstacle by only using a 2D bounding box with its centroid given and the low-cost image acquisition devices (webcams). Working in the hue-saturation-value (HSV) color space, it can differentiate the challenging obstacle by color and lighting conditions. A key feature is the use of two independent cameras to avoid the obstruction of vision by the robotic arm, as it invades the vision field of view. Therefore, the system relies on a main overhead camera (cam-1), a support lateral camera (cam-2), and an efficient switching strategy proposed for a dynamic shift between cameras. Another key feature is that the HSV parameters can be easily adjusted offline by a real-time visualization tool. Preliminary experiments and live demonstrations show that a relatively robust performance for object detection and tracking integrated with robot control and path planning is satisfactory, even in challenging and dynamic conditions such as non-uniform background, robot interference and vision obstruction. The proposed system can be further enhanced by exploiting more efficient image processing algorithms at feature level.
Period23 Apr 2018
Event title9th China-Scotland Signal and Image Processing Research Academy Workshop
Event typeWorkshop
LocationSuzhou, China
Degree of RecognitionInternational