Real-Time Monitoring and Detection of Defective Pins

Teng, Wei Kit (2024) Real-Time Monitoring and Detection of Defective Pins. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text TENG WEI KIT_Full Text.pdf Restricted to Registered users only Download (10MB)

Abstract

There are many real-time approaches to performing LED Pins detection. One of the most renowned methods for real-time LED Pins detection is the CNN approach. CNN approaches have a large family, which can be divided into single-stage detectors and double-stage detectors. The two-stage object detectors such as R-CNN, Fast R-CNN, Faster R-CNN, and Mask R-CNN typically achieve the highest accuracy in object detection. However, they are time-consuming and unsuitable for a real-time LED Pins detector, as the frames per second (FPS) are lower than those of single-stage object detection due to multiple inference steps per second. The single-stage detectors such as SSD and YOLO require only a single pass through the neural network to perform object detection, making them suitable for real-time LED Pins detection. Although SSD has better accuracy than the YOLO method, it struggles with recognizing smaller objects. For this research, the YOLO family, particularly YOLOv8, produces very high accuracy for object detection with a very fast response time. Therefore, YOLOv8 was used in this project because of its real-time and stable accuracy performance. YOLOv8 is a popular object detection algorithm, widely used for its high precision and real-time processing capabilities. However, due to its complex neural network architecture and the substantial computations required for object detection, YOLOv8 requires powerful GPUs for efficient operation. Anaconda, an online GPU service, provides access to high-performance GPUs, despite limitations such as time constraints, memory and storage space, and network latency. While YOLOv8 exhibits high precision and real-time processing capabilities, detecting small objects and complex scenes in images or videos remains a challenge. Therefore, this algorithm was trained and evaluated using the LED Pins detection dataset. In this project, a real-time monitoring and detection system was developed for identifying defective pins using the YOLOv8 algorithm trained on a custom dataset. Through rigorous experimentation, parameters and preprocessing techniques were optimized, achieving superior performance compared to YOLOv4. This detection system successfully achieved a detection accuracy of 98% precision and 99.5% mean average precision at an IoU threshold of 0.5 on a test dataset of 12,000 LED pins. Additionally, it has integrated real-time monitoring capabilities, providing manufacturers with proactive quality control measures and enhancing overall product quality. This system represents a significant advancement in quality control within the manufacturing industry, with the potential to revolutionize quality assurance processes and enhance product reliability.