Robot Vision course projects | NYU Tandon School of Engineering
Implementing visual place recognition, maze navigation, and core computer vision algorithms from scratch.
vis_nav_player/
├── midterm/ # Visual maze navigation system using DINOv2 + FAISS
├── final_project/ # Full autonomous maze solver with visual SLAM concepts
└── colabs/ # Weekly hands-on implementations (Labs 01–11)
├── 02_Intro_to_DL.ipynb
├── 03_Camera_Calibration/
├── 04_Tag_Based_AR/
├── 05_Stereo_Hands_On/
├── 06_SfM.ipynb
├── 07_ICP/
├── 08_RANSAC/
├── 09_Detection/
├── 10_Tracking/
└── 11_Segmentation/
Each folder contains my implementation of the weekly hands-on session, submitted as coursework for ROB-GY 6203 — Robot Vision at NYU.
| Lab | Topic | Key Techniques |
|---|---|---|
| 02 | Intro to Deep Learning | Neural networks, PyTorch basics |
| 03 | Camera Calibration | Intrinsic/extrinsic parameters, chessboard calibration |
| 04 | Tag-Based AR | AprilTag detection, pose estimation |
| 05 | Stereo Vision | Disparity maps, depth estimation |
| 06 | Structure from Motion | Feature matching, 3D reconstruction |
| 07 | ICP | Iterative Closest Point, point cloud alignment |
| 08 | RANSAC | Robust homography estimation |
| 09 | Object Detection | Detection pipelines |
| 10 | Tracking | Multi-object tracking |
| 11 | Segmentation | Semantic/instance segmentation |
Built a visual navigation system that guides a robot through a maze using only camera images — no odometry, no GPS.
Key tech: DINOv2, FAISS, OpenCV, Pygame
Extended the midterm system to achieve full autonomous maze traversal with real-time localization and path planning from purely visual inputs.
Key tech: VLAD, ORB, FAISS, SIFT, OpenCV, NumPy
Pedro Felix
Electrical Engineering | NYU Tandon School of Engineering
GitHub