Mobile Device-Based Ants Recognition and Tracking System: Methodology and Frameworks

This research introduces a practical framework for ex situ ants recognition and tracking, enabling the continuous monitoring of their movements.This framework includes an iOS-based client application with an embedded Visual Intersection Over a Union (V-IOU) tracking module and a You Only Look Once (YOLO) recognition model.Another part of a framework is a scalable system for autonomous annotating, training, and converting the model to mobile format.The tracking algorithm is integrated into a Swift application using the JavaScriptCore library, while the Yolo model is integrated using the CoreML framework.To ensure system accuracy and stability, the method of improved clasterization K-means++ is employed.Simultaneously, the Affine Quantization method is used keep the model size as small as possible.An experimental benchmark on an indoor ant colony was conducted, using various recognition models to assess the accuracy and productivity of the system.The results confirm the practicality of our methods and frameworks for real-time small object detection, demonstrating their applicability in real-world scenarios.