Mobile Visual Classification (MVISC) is a project to identify and classify animals. The goal is to provide: Computer Vision Based Individual Animal Recognition.
MVISC is a software that lets you take photos of animals and analyze their unique visual patterns. Within minutes, the software is able to tell you, whether an animal of a certain species, has already been found, or if it is a new individual. Thanks to state-of-the-art computer vision algorithms, these assertions come with a very high accuracy, namely around 90-100%!
Visual recognition is highly efficient, that means no more invasive marking methods are required - stress may be reduced, drastically. We intend to offer the app as an online and offline solution. For offline use cases, where no Internet is available we are also working on hardware using standard components.
Please join chat channel at https://gitter.im/fossasia/mvisc
The project will be shipped along with the software, that will hold the following key functions:
- Take a photo of an individual animal, including fully automatic image quality check
- Run analysis and compare new individual with individuals in the local database
- Let the software assert, whether the individual is new or already exists (match)
- Manage database entries, add descriptions and export
How does it work? We make use of modern approaches from the field of computer vision, including image description and clustering algorithms. The result so far already delivers outstanding classification accuracies. For large-scale animal databases, machine learning algorithms may be utilized.
Unlike the figure suggests, the photos do not necessarily have to be aligned or cropped - the software does it! Photos are being pre-processed, and the description of patterns is rotational invariant. The only thing you have to do is to take the picture.
The project is based on the former Zooracle project and licensed under GNU General Public License v3.0.