Development of real-time multi pose face recognition and tracking system

• Main Author: Manimehala Nadarajan
• Format: Thesis
• Language: English; English
• Published: 2016
• Subjects: TA1501-1820 Applied optics. Photonics
• Online Access: https://eprints.ums.edu.my/id/eprint/39297/1/24%20PAGES.pdf; https://eprints.ums.edu.my/id/eprint/39297/2/FULLTEXT.pdf; https://eprints.ums.edu.my/id/eprint/39297/

The birth of telepresence robots in healthcare industries has made a significant transformation in the last few decades. Telepresence robot is remotely connected and embodied to perform several task such as patient monitoring, diagnosis, surgery and other task. Due to poor infrastructure especially in the interior of developing countries, a Medical Telediagnosis Robot (MTR) which works with a low bandwidth and on a low cost platform was developed. Unlike other tele-presence robots, MTR is capable of performing remote diagnosis during medical emergencies as it is equipped with basic medical instruments and dual vision system which comprises of a visual diagnostic system and a visual communication system. Visual communication system in MTR provides a basic face-to-face communication. The application of biometric system using face can greatly improve the current visual communication system as it is currently limited for only manual face recognition and tracking. It is difficult for the remote medical specialist to keep the patient and medical staff in an ideal field of view (FOV). It is also necessary for the remote medical specialist to identify the correct patient and medical staffs for diagnosis and verbal communication. To circumvent this problem, a real time face detection, recognition and tracking system (DRiT) is developed. To achieve a real time system, the DRiT system is designed with four modules which are operated in sequence and thus minimizing the execution time. Other challenges that were circumvented by the DRiT system are multi face pose, varying background condition during camera movement and changes in environment lighting with respect to time. DRiT is fully designed in LabVIEW platform which integrates software, hardware and GUI modules to complement with the current MTR platform. The background, lighting conditions and face pose were solved using hybrid approach utilizing skin color information to detect face. Neural Network was deployed to identify the profile of a person in multi poses and distances. A hardware together with software based face tracking is designed to ensure that the face region is still within the tracking view. Tracking a person continuously in a wider angle is a challenging task but this has been successfully achieved with DRiT system using a pan and tilt unit. DRiT is a standalone platform which is activated once the robot is navigated to the desired area. DRiT system creates a better visual communication between remote specialist and hospital members as the remote medical specialist will no longer require to execute manual control of the robot. The developed DRiT system was experimentally tested in real time and it yields an accuracy of 98% with an execution time of 56 ms.