Mosaic creation through the stitching of aerial rectified images obtained from a moving vehicle

Goal

Mosaic creation through the stitching of aerial rectified images obtained from a moving vehicle.

Dataset

Stuttgart of Kitti (https://cityscapes-dataset.com/downloads/), in particular leftImg8bit_demoVideo (https://drive.google.com/open?id=1GrhTcMOxjPN-Ei-k9IoUW1TAGNXe-ktn). You can put the frames to processing into the folder dataset. To compute correctly the rectified images, use the right camera info into the MATLAB script (camera.json for stuttgart https://drive.google.com/open?id=1tDMvRqau5c_hhBb1UioJgK9GP6kyJfgM).

Pipeline

  1. Cars detection
  2. Rectification
  3. Keypoints detections with SURF
  4. Compute matrix Homography
  5. Image warping
  6. Overlap to mosaic creation

To run

$ python3 car_detection.py

To compute the rectified images run on MATLAB the script:

compute_bird_view.m
$ python3 create_submosaic.py
$ python3 merge_submosaic.py

Submosaics

Results

First submosaic Second submosaic

Merge submosaics

Results:

First submosaic + second submosaic

Requirements

Software Version Required
Python >= 3.5 Yes
Numpy Tested on 1.13 Yes
Matplotlib >= 1.0 Yes
opencv-python == 3.4.2.17 Yes
opencv-contrib-python == 3.4.2.17 Yes
os - Yes