Description

AI Anomaly Generation & Machine Learning Validation
Project Title
Computer Vision & AI Framework for Automated Video Anomaly Generation and ADL Recognition
General Description
Detecting functional decline in elderly individuals through passive video monitoring requires AI models capable of identifying errors, omissions, or unsafe behaviors during daily tasks. A major obstacle in this domain is extreme data scarcity, as real-world video recordings of elderly people making errors are rare, costly to collect, and ethically sensitive. To overcome this, this project focuses on developing an AI framework that takes normal execution videos of household activities, automatically breaks them down into temporal phases, and synthetically generates realistic structural and visual anomalies. The validity of this augmented dataset will then be proven by training and evaluating a dedicated computer vision model.
Project Scope
The main objective of this project is to develop an algorithmic computer vision pipeline for automated video anomaly generation and machine learning validation. Students will implement temporal action segmentation to split continuous activity videos into key operational phases. They will create an automated anomaly generator utilizing both temporal augmentations (e.g., phase omission, sequence shuffling) and visual augmentations (e.g., inpainting objects/actions), along with an automatic annotation system producing standardized metadata (JSON/YOLO/COCO). Finally, students will validate the generated dataset by training a machine learning/video classification model on a primary anchor action (e.g., boiling water in a kettle), evaluating its performance using standard metrics (Precision, Recall, F1-Score, ROC-AUC), and testing the framework's generalization across 2–4 additional household activities.
Student Requirements
●	Teamwork and active collaboration on complex algorithmic tasks
●	Full attendance in weekly advisory meetings
●	High motivation and strong interest in Computer Vision and Deep Learning
●	Independent learning capabilities to explore state-of-the-art vision models and augmentation libraries
●	Personal responsibility for experimental rigor, model training, and codebase documentation
Development Tools
●	Programming Language: Python
●	Deep Learning Frameworks: PyTorch
●	Computer Vision & Video Processing: OpenCV, ImageIO, MoviePy
●	Segmentation & Augmentation Models: Pre-trained Temporal Action Segmentation models, Inpainting models (e.g., Stable Diffusion / LaMa)
●	ML Evaluation & Experiment Tracking: Scikit-Learn, Pandas, Matplotlib/Seaborn, WandB / MLflow
●	Version Control & DevOps: Git, GitHub
Deliverables
●	Specification document
●	Video
●	Poster
●	Presentation
●	GitHub link to the code
For inquiries:
Ido Freidlin | Mobile: 054-5619201
AI Anomaly Generation & Machine Learning Validation

AI and Machine Learning

Description

Emphasis in project execution

The project is has cooperation with the industry and combines meeting deadlines while being creative and focused on the task
