Radish AB - Adjustment of facial recognition system to enable client-based predictions - Prestudy

Purpose and goal

The purpose and goal of the project was to improve the company´s competence and understanding for the possibilities for computer vision analysis on cellular devices, and improve the current system. This has been achieved very well. The technology has been a contributing factor in helping the company attract investors and complete the company´s first financing round, which was completed at the end of the project-period. The company is deemed to have good possibilities to continue cooperation with parties from Chalmers University of Technology.

Results and expected effects

The project has resulted in that the company was able to move parts of the recognition system from the cloud to work locally on the phone. New knowledge of shape regression has resulted in new software for live analysis of specific facial points. This can be used to determine position, rotation and facial expression. The result is therefore even better than the company was hoping for in the beginning of the project.

Approach and implementation

We started by increasing our knowledge of the field. Over 40 papers where studied, as well as meetings with representatives from Chalmers algorithm department. Thereafter the company was able to serialize the PCA-matrix and thereby succesfully complete analysis on a mobile phone. We were looking for a way to define the rotation of a face, since it is important for face recognition. We decided to explorer if shape regression could be performed on a phone. We deeped our knowledge in this field, and after some time managed to complete software to be run on a mobile phone.