This repository contains a semester's worth of computer vision research including much related to automatic "clutter" detection. In this case, clutter consists of dense clusters of small objects in scenes.

A large and complex component of the Python module presented here is the main tool, which provides a common entry point for a number of vision routines. To view its help text and options, use python -m vision --help from the root of this repository.

Not all possible uses of the tool are functional or documented.

In general, a detection tool may be invoked as

python -m vision [routine] --overlay --draw [input_images...]

This will run the detector routine on each image, overlay the result mask on the original, and draw the image onscreen.

Notably, the routines used for the report in this project are

python -m vision corners --overlay --draw data/first_pass/*
python -m vision contour_dense_edges --overlay --draw data/first_pass/*

Overlaid results may be colored with --colorize [red|blue|green|white].

They may be saved with --save [path/to/output/directory], and each image will retain its basename.

To prevent overlaying the results, omit the --overlay flag.

If ground truth layers are available, supply -T or --truth.

The vision module is written in Python 2, and requires OpenCV 2 or 3, Numpy, and imutils. If OpenCV is already installed on your computer with Python 2 bindings, a virtual environment with the required libraries may be prepared with make env.