ExypnoTech Engineering Services - EXY_02 Develop an intelligent system for real-time analysis of fish behavior using underwater 2D/3D vision; detect and track fish movements; extract key behavioral indicators such as speed, agitation, grouping, and vertical position; compute a Stress Score reflecting fish welfare; integrate Deep Learning, multi- object tracking, and depth processing into a unified pipeline; display stress levels, heatmaps, and abnormal behavior alerts through a real-time dashboard; and validate the system using video data from offshore cages and RAS tanks. Real-Time Fish Behavior Analysis & Stress Detection Using Deep Learning and Underwater 3D Vision PFE | Hi Interns

Description du poste

Project overview

Develop an intelligent, unified pipeline for real-time analysis of fish behavior using underwater 2D/3D vision.
Target outputs include detection and tracking of fish, extraction of behavioral indicators (speed, agitation, grouping, vertical position) and computation of a Stress Score to reflect fish welfare.

Set up underwater video acquisition and collect/annotate behavior datasets for both offshore cages and RAS (Recirculating Aquaculture Systems) tanks.
Implement a detection + tracking pipeline capable of real-time monitoring; integrate depth processing to support 3D position/vertical position analysis.
Extract behavioral features, generate movement heatmaps and behavioral statistics, and develop a Stress Score model to quantify welfare.

Validate the system using video data from real offshore cages and controlled experimental RAS tank conditions.
Build a real-time dashboard showing stress levels, heatmaps, and abnormal behavior alerts; prepare documentation, experiment reports, and technical presentations.

Apply Computer Vision & Deep Learning fundamentals to design and implement detection and tracking modules.
Work in Python using OpenCV and deep learning frameworks such as PyTorch or TensorFlow; implement object detection models (examples given: YOLOv8 / YOLOv11) and multi-object trackers (StrongSORT, DeepSORT, ByteTrack).
Understand machine learning basics for feature extraction and Stress Score model development; be prepared to produce experiment reports and presentations.

A real-time pipeline which detects, tracks and computes behavioral metrics for fish, produces movement heatmaps and a Stress Score, and raises alerts for abnormal behavior.
Comprehensive documentation, annotated datasets, experiment reports, and a working dashboard validated on both offshore and RAS video datasets.

Preferred tools: Python, OpenCV, PyTorch or TensorFlow; object detection models such as YOLOv8/YOLOv11; trackers such as StrongSORT, DeepSORT, ByteTrack.
Additional tasks: dataset annotation, depth processing for 3D/vertical position, generation of behavioral statistics and visualization (heatmaps, dashboards).