DuoCounter An Approach for Counting Fish Using Tracking

DuoCounter is a web-based system developed for fish detection and counting using object detection and tracking techniques. The system uses two modern detection models, YOLOv26 and RT-DETR, trained on the Fish Detection (Labelled) dataset containing 13 fish species. After detection, the system applies the BoT-SORT tracking algorithm to assign unique identifiers to each fish across video frames. The counting is performed by maintaining a master set of unique IDs. Once a fish receives a new ID, it is registered and counted only once. This method removes the need for virtual lines or complex re-identification steps used in earlier systems. The complete pipeline includes detection, tracking, confidence filtering, and final counting. A web application built using the Flask framework provides multiple features. Users can register and login to access the system. The application supports image upload for species detection and video upload for automatic fish counting. Results are shown with bounding boxes, species names, unique track IDs, and confidence scores. An annotated video output is also generated for review. The system aims to provide an accurate and easy-to-use solution for fish monitoring. Experiments conducted on the Fish Detection (Labelled) dataset show good detection performance and reduced counting errors.

Keywords: DuoCounter, Fish Detection, Fish Counting, YOLOv26, RT-DETR, BoT-SORT, Unique ID Tracking, Flask Framework, Object Detection, Multi-Object Tracking.