3liYounis/Theft-Detection-System
Computer Vision & Deep Learning models for tracking user behavior in shops, predicting real-time probability of shoplifting and enables instant alerts to a mobile app. The full report includes information about the thief extracted by the facial encoding and with the help of Face Recognition pre-built model.
What's novel
Computer Vision & Deep Learning models for tracking user behavior in shops, predicting real-time probability of shoplifting and enables instant alerts to a mobile app. The full report includes information about the thief extracted by the facial encoding and with the help of Face Recognition pre-built model.
Code Analysis
9 files read · 3 roundsA real-time video surveillance pipeline that uses YOLOv8 for person detection, MediaPipe for pose extraction, a 3-layer LSTM for temporal theft behavior classification, and InsightFace for suspect identification, with Firebase alerting.
Strengths
The feature engineering is thoughtful and domain-appropriate (torso-normalized joint positions, hand-to-hip distances, elbow velocity/acceleration), and the training pipeline is well-structured with proper data splits, class weighting, and comprehensive evaluation metrics.
Weaknesses
Critical bugs exist (video_path never passed to pipeline, per-person model loading is a performance disaster, missing Firebase service crashes the alert module), there are no meaningful tests, and the code has duplicate blocks and inconsistent path handling that suggest it was assembled quickly without integration testing.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
58
Contributors
3
Files
26
Active weeks
14
Repository
Language
Python
Stars
1
Forks
0
License
—