Obstacle detection based on the optical flow

This algorithm is based on the paper: N. Ohnishi and A. Imiya, “Appearance-based navigation and homing for autonomous mobile robot,” Image and Vision Computing, vol. 31, no. 6, pp. 511–532, 2013.

Link to paper

Installation

OS: Ubuntu 18.04

Requirements:

1. Python 2.7
2. OpenCV (tested on version 3.3, older version should work as well)

Options

The obstacle detector contains three option for key points selection: 1) Harris corner detector, 2) Harris corner detector per region, 3) Vertices of mesh.

There are two option to compute the planar flow: 1) Homography, 2) Affine transformation.

Usage

You can test the obstacle detector adding your image sequence in a folder called "images" inside the Obstacle-Detector-Optical-Flow root directory.

Then run

cd [Obstacle-Detector-Optical-Flow root dir]
python optical_flow_detector.py

To test diferent key points selection, edit the variable key_points_selector in the script.

Result

Optical Flow:

Planar Flow:

Obstacle key points (RED):

Bounding boxes:

Contact:myrna.castillo.silva@gmail.com