This repository contains computer vision projects focused on motion estimation and 3D scene reconstruction from images and videos. The work emphasizes geometric understanding and optimization-based perception pipelines commonly used in robotics.
The primary objective of these projects is to build a strong foundation in vision-based perception systems, rather than end-to-end black-box learning. The implementations focus on understanding how motion and 3D structure can be inferred from visual data.
- Estimated dense optical flow from video sequences
- Analyzed pixel-wise motion between consecutive frames
- Based on classical assumptions such as brightness constancy and spatial smoothness
- Useful for understanding motion estimation and visual odometry foundations
- Reconstructed 3D structure using multiple camera views
- Applied pinhole camera models and multi-view geometry concepts
- Performed triangulation and basic pose estimation
- Relevant for depth perception and scene understanding
- Classical computer vision methods
- Geometric modeling and camera projections
- Linear and nonlinear optimization concepts
- Emphasis on interpretability and system-level understanding
- Python
- OpenCV
- NumPy
- Jupyter Notebooks / Google Colab
These projects were implemented to build a strong foundation in vision-based perception systems, particularly for robotics applications such as SLAM (Simultaneous Localization and Mapping), navigation, and state estimation.
This work was completed as part of the selection tasks for the Software Team of the Autonomous Ground Vehicle (AGV) project at IIT Kharagpur.
This repository is a work in progress. Future improvements may include better evaluation metrics, visualization of results, and performance optimizations.