Physics-aware vision instrumentation for stingless bee counting at hive entrance using hybrid edge-cloud object detection

¹ Faculty of Electrical Engineering, Universiti Teknologi Malaysia, 81310 Johor Bahru, Johor, Malaysia
² Faculty of Mathematics and Natural Sciences, Universitas Negeri Jakarta, 13220 Jakarta Timur, Jakarta, Indonesia
³ Departemen of Electrical Engineering, Institut Teknologi Indonesia, 15314 Tangerang Selatan, Banten, Indonesia
⁴ Department of Chemical Engineering, Politeknik Negeri Bandung, 40559 Bandung Barat, Indonesia
⁵ Research Center for Smart Mechatronics - National Research and Innovation Agency (BRIN), Indonesia

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 2 July 2026

Abstract

Stingless bee entrance monitoring requires a non-invasive tool to measure colony traffic without disrupting foraging. A hybrid edge-cloud object identification system and physics-aware vision instrumentation framework are used to track stingless bees in this paper. The system uses a Raspberry Pi 4 edge node, Sony IMX296 Global Shutter camera, Only Look Once (YOLO)-based detection, and Observation-Centric Simple Online and Realtime Tracking (OC-SORT) tracking. Because entry activity implies foraging intensity, the traffic count can be a non-invasive proxy for colony health and yield. The edge-side YOLO11 and OC-SORT tracking pipeline found trajectory fragmentation during high-speed ingress. A frame interval of 0.0667 s was achieved using a 15 FPS edge processing rate. Inter-frame displacement may approach 0.333 m for bee motion exceeding 5 ms^-1, producing missed detections and identity switching. Thus, a sampling-based tracking failure situation happens when a bee travels more than the tracker's maximum association distance between two processed frames. This illustrates that frame rate, bee velocity, field-of-view scale, detector recall, tracker association tolerance, and biologically meaningful bee counting interpretation all affect bee tracking reliability. Validation using 14 one-minute Geniotrigona thoracica entrance videos showed an average IN counting accuracy of 85.0%, OUT counting accuracy of 66.7%, and total counting accuracy of 76.3% compared with manual video counting.