# IMFOA-OFT: Semantic and Optical-Flow SLAM for Dynamic Indoor Environments

## Description
This project implements a dynamic monocular SLAM system using improved Mask R-CNN with CFPN for dynamic object segmentation and Elite Moth–Flame Optimization Algorithm (EMFOA) for feature tracking via optical flow.

## Dataset Information
- **Dataset**: TUM RGB-D Dataset  
- **URL**: [https://vision.in.tum.de/data/datasets/rgbd-dataset](https://vision.in.tum.de/data/datasets/rgbd-dataset)  
- **Scenes**: rgbd_dataset_freiburg3_sitting_xyz, walking_xyz, etc.

## Code Information
- `DynamicSLAM`: Full SLAM pipeline.
- `ImprovedMaskRCNN`: Semantic segmentation using Mask R-CNN + CFPN.
- `OpticalFlowTracker`: Feature tracking using optical flow and EMFOA.
- `EMFOA`: Metaheuristic optimization algorithm.
- Dependencies include: `torch`, `opencv-python`, `scikit-learn`, `numpy`.

## Usage Instructions
```bash
# Install dependencies
pip install -r requirements.txt

# Run the SLAM system on a video
python run_slam.py --input dynamic_scene.mp4
```

## Requirements
- Python 3.8+
- Required libraries: numpy, opencv-python, torch, torchvision, scikit-learn

## Methodology
- **Segmentation**: Improved Mask R-CNN with CFPN for dynamic object detection.
- **Tracking**: EMFOA-optimized optical flow.
- **Geometric validation**: Fundamental matrix estimation with polar constraint.
- **Evaluation**: TUM RGB-D benchmark using RPE and APE metrics.

## Citations
> Chen et al. (2022). A Monocular-visual SLAM system with semantic and optical-flow fusion for indoor dynamic environments. *Micromachines*, 13(11), 2006.

## License & Contribution Guidelines
This project is licensed under the MIT License. Contributions are welcome via pull requests.