Collaborative Cloud-based FR Approach for Humanoid robots

The ability to recognize human faces in real time is an important requirement for most humanoid robots. One of the challenges in face recognition (FR) applications is the time it takes a robot to search through a large dataset of known faces. As a database of known images is increasing, a robots ability to store and process the data in real time is decreasing. In this paper we present a new cloud-based FR algorithm which will enable faster processing of data in face detection and recognition by humanoid robots. An improvement in robots performance will be achieved by offloading storage and processing tasks from limited on-board robot resources to the resources in the cloud. In the case of multi-robot systems, their performance can be further improved through cloud-based collaborative learning and information sharing. We created a new dataset containing over 300 trained facial images of 10 people. The result shows that the proposed FR can achieve 83% accuracy rate and exceeds the local FR performance in terms of the response time which is slightly increased in comparison to local system performance which sees significant increase when the dataset size grows. The system proved its capability to share knowledge between robots in the same multi-robot environment.