A classicial computer vision based approach to transform images from multi-camera sensors to a single bird's-eye view (BEV) image. It provides a simple implementation for performing bird's-eye view (BEV) transformation using homography. It involves selecting points manually on the images, computing homographies, and transforming the images to a bird's-eye view perspective.
- Load and resize images.
- Manually select points on images for the BEV transformation.
- Compute homographies using selected points.
- Warp images to the bird's-eye view perspective.
- Stitch the warped images together to create a final BEV image.
Run the following command form the root directory:
python main.py
- The
selected_points.jsonfile is used to save the points you manually select. If you already have this file, the script skips the manual selection and use the saved points. - The images should be correctly oriented and the selected points should correspond to the desired region for the bird's-eye view.