Real-Time Fish Detection and Tracking using Ensemble based Deep Learning Approach for Aquatic Species Monitoring

This study presents an advanced deep learning framework for real-time fish detection, classification, and motion tracking in underwater environments. YOLOv8 is utilized for fast and accurate fish detection, while motion-based keyframe extraction and optical flow algorithms like Lucas-Kanade and Farneback enhance tracking performance. ByteTrack ensures consistent object tracking across frames, and classification models, including SVM, Random Forest, and CNN, improve species identification. To optimize real-time performance, the system applies preprocessing techniques such as Gaussian blurring and contrast enhancement, along with post-processing methods like Non-Maximum Suppression (NMS) to refine detections. Designed for live underwater monitoring, this approach enhances fish behavior analysis, ecological research, and fisheries management.