Design and development of an image-guided vision system for robotics palletizing / Mohamad Zaid Mohamad Zaihirain
The ever-evolving trend in modern manufacturing techniques has led to a shift from the conventional method to the heavily automated manufacturing process. Modern intelligent technologies are being used to simplify, accelerate, and improve the quality of traditional manufacturing methods. Automation...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/13435/1/Mohamad_Zaid_Mohd_Zaihirain.jpg http://studentsrepo.um.edu.my/13435/8/zaid.pdf http://studentsrepo.um.edu.my/13435/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The ever-evolving trend in modern manufacturing techniques has led to a shift from the conventional method to the heavily automated manufacturing process. Modern intelligent technologies are being used to simplify, accelerate, and improve the quality of traditional manufacturing methods. Automation of production lines through robotic implementation can improve the manufacturing performance while lowering the associated costs because it helps standardize the stacking and palletization procedures. Current automation on palletizing system relies heavily on a predetermined sorting system due to its inability to detect irregular-shaped object. A vision system is needed to increase the flexibility of the robotic palletizing system by detecting the type of object and its orientation. To this end, this project aims to design and develop an image-guided vision system through the application of YOLO object detection and OpenCV for small-scale robotics palletizing of non-uniform shaped object, i.e., 3D printed chicken wings and drumsticks. A YOLO object detection model is trained using 5000 images containing the chicken wings and drumstick. This object detection model is then used to determine the type of object. Then, this information on type of object is used along with an orientation detection program to find each object�s orientation. The orientation detector is programmed with a contour detector called �Canny Edge detector� and a �fitEllipse� function that generates the angle of orientation. Using the location and orientation information generated by the detection programs, a pick and place operation is simulated in RoboDK. Through several case studies, this detection models works great if the objects in the images are arranged in a certain way, i.e., not closely packed together, or overlaps onto one another. In an ideal case, the pick and place simulation work flawlessly with the information obtained from the YOLO object detection and orientation detection program |
|---|