Publication: Image classification and 3d coordination on multiple objects
Loading...
Date
2023-07
Authors
Lim, Boon Hou
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The Fourth Industrial Revolution refers to the increasing automation and data exchange in manufacturing processes, incorporating technologies like
cyber-physical systems, IoT, cloud computing, cognitive computing, and artificial intelligence. Within this concept, industrial automation plays a
significant role by automating industrial processes and machines. Humanoid robots, designed to mimic human movements and behavior, offer improved
speed and accuracy in performing tasks. In this project, the focus is on developing a humanoid vision system capable of object recognition and 3D
coordination. The system utilizes the Intel RealSense Depth Camera D435i to capture depth information from the surroundings and employs a Python
program for data interpretation. A custom YOLOv5 model is trained using images from the humanoid's working environment. The developed vision system successfully detects targeted objects in the surroundings and provides depth information for specific points. The accuracy of the distance
measurement is evaluated, demonstrating an average percentage error of 6% to 7% for items within a 2-meter range. Evaluation of the trained model reveals
a precision, recall, and F1 score of 92.81%, indicating that the model meets the required 90% accuracy threshold and is ready for deployment. In summary,
this project achieves success as the developed system performs both object detection and distance measurement with high accuracy.