A real-time computer vision system that predicts object motion intent — MOVING_AWAY, CROSSING, APPROACHING, and COLLISION_RISK using optical flow, temporal feature extraction, and lightweight machine learning.
🚀 Why this project? Most vision systems detect objects. This system goes one step further by predicting motion intent, enabling early collision-risk detection — a key requirement in autonomous drones and robotics.
🧠 System Overview
- Real-time webcam input
- Motion detection via frame differencing
- Optical flow for direction and speed
- Interpretable feature extraction
- Logistic Regression for intent classification
- Scene-level decision aggregation for stable output
📊 Output
- Object-level bounding boxes with intent labels
- Single scene-level conclusion displayed on screen
- Collision risk highlighted clearly
🛠️ Tech Stack
- Python
- OpenCV
- NumPy
- scikit-learn