Real-Time Pig Counting Embedded System via Video Object Detection and Tracking

Pig farming currently plays a significant role in providing food for humans. However, as farms grow larger, exporting becomes increasingly challenging, especially in terms of accurately counting the number of pigs. With manual counting methods, the rising number of pigs leads to human errors, increased labor costs, and long counting times that can stall the export process. Experimental results show that even farm owners struggle to control the export quantities due to limitations in statistical tracking and verification. In this study, we develop a real-time pig counting embedded system that uses object detection and tracking from a camera in the counting zone to overcome inefficiencies in traditional pig-counting methods on farms. Our system also provides a simple web application to store pig counting results including quantities and videos, provides statistical charts to visualize those data. Our system uses YOLOv5 for object detection and a novel IndexTrack algorithm for object tracking, which was designed specifically for deployment on an NVIDIA Jetson Xavier NX embedded board. This system directly solves the problems of human counting errors and long counting times in manual methods. Our experiment results show that the system provides a lightweight method that is capable of functioning in a real-world environment and is effective in terms of both accuracy and execution time (real-time execution), even when some pigs pass through the counting zone back and forth.