Task 5 Feature Descriptors

[rgolovanov]

Feature Descriptors

Description

The main goal of the task is to meet on practice how feature descriptors are constructed.
In this particular task students should implement their own binary descriptor to provide "good" distinctness. This task is a bit research:

• implement binary descriptor algorithm,
• use demo_feature_descriptor to dump your binary descriptor and any standard one,
• compare distances between your descriptors and standard ones and visualize it in histogram with normalized range [0, 1],

Workflow

1. Fork current repository to your own public one or update yours with new commits from this one.
2. Provide solution in your repository (resolve all \todo in the code).
3. Create "Pull request" in your repository and mention lecturer to receive online feedback.
4. Show complete solution on the next (or current) lesson in computer classroom.

Initial state

Initial code contains some dummy descriptor generation based on rand() function. Demo application for lab 5 looks similar to lab 4, but provide descriptor calculation for detected corners. For each corner we computes dummy descriptors and some standard (ORB). Then all these descriptors are stored into descriptor.json.

Special helper python's script is provided to plot histogram of distances for each type of descriptor.
The more to the right is the peak, the better the quality of the descriptor:

build/ $ ./demo/Release/cvlib_demo.exe

[5-> press space to create file with descriptors]

build/ $ python3.exe ../demo/demo_feature_descriptor.py descriptors.json