Image Stitching for Drone Mapping

This project contains code for stitching large number of images using classical computer vision techniques for application in drone mapping.

Setup python environment

Install conda and prepare a python environment using the following steps:

Build a new conda environment

conda create -n image_stitching python=3.8

Install additional libraries using pip:

python3 -m pip install -r requirements.txt

Activate the conda environment:

conda activate image_stitching

Approach

Assumptions

• There is atleast 70% overlap (both front and sideways) between the images while the data was collected
• The images are numbered sequentially in the dataset

Methodology

Below is a high-level overview of the image stitching algorithm:

1. Feature detection using SIFT
2. Feature matching using FLANN-basd matcher
3. Homography estimation using DLT and RANSAC
4. Image warping

Stitched Mosaics

The algorithm is tested on two datasets: agricultural farm and an urban area. Below images shows two examples mosaics:

Farm Mosaic

City Mosaic

Future Work

Two possible improvements over our work is to use bundle adjustment to reduce the cumulative error issue and blending to remove noticeable seams at the stitching location.