Human-AI Collaboration - Impact of Dynamics of Access and Usage

Supervisors: Samy Rima, Simon Ruffieux, Denis Lalanne

Contact person: Samy Rima

Student: Looking for student

Project status: Open

Year: 2025

This project aims to develop an integrated system for automated macaque identification and tracking in the lab of the University of Fribourg, building on recent advances in detection, pose estimation, face recognition, and multi-frame tracking.

The student(s) will implement a full pipeline that ingests video or live camera feeds, detects macaques using deep learning models (e.g. Mask R-CNN with transformer backbones), estimates body pose from keypoints (leveraging datasets like MacaquePose), recognizes individual identities through face recognition models trained on annotated images, and links detections across frames with tracking algorithms such as Kalman filters and Hungarian matching to maintain consistent trajectories. The system should also support data acquisition and annotation, balancing manual labeling with semi-automated approaches, and provide robust evaluation metrics for detection accuracy, pose error, identity recognition, and tracking stability. Note that some of the pipeline components can leverage past research such as (OpenMonkeyStudio)[https://github.com/OpenMonkeyStudio] or (MacqD)[https://github.com/C-Poirier-Lab/MacqD].

A key deliverable is a user interface that allows researchers to load videos, run analyses, visualize overlays (bounding boxes, pose skeletons, identity labels, motion paths), manually correct misclassifications, and export results for downstream analysis of behaviour and welfare. Student(s) will progress through milestones including dataset preparation, model training, pipeline integration, and UI prototyping, with optional extensions into 3D pose, behaviour classification, and real-time analysis. The final deliverables will include a working prototype with documented code, annotated datasets, trained models, performance evaluation, and a functional UI, directly enabling non-expert lab members to monitor macaque activity and identity in a transparent, user-friendly way.

Interesting references:

• https://pubmed.ncbi.nlm.nih.gov/28739161/
• https://www.nature.com/articles/s41467-020-18441-5
• https://www.nature.com/articles/s41598-025-95180-x
• https://pmc.ncbi.nlm.nih.gov/articles/PMC9113979/

Keywords: Machine Learning, Deep Learning, Computer Vision, Dashboard, Data Analysis.

Document: Not yet available