Eye on Target: Embedded Robotic Emulation of Mammalian Gaze Fixation Using Computer Vision

When an object caught the attention of an mammal, the mammal will fix its gaze on that object. This might involves turning the head and/or rotate the eyeballs. The same behavior is also desirable in numerous robotic applications. With the advances in computer vision and micro-controller technologies, it is possible to empower a robotic project with this kind of object tracking capability. You Only Look Once (YOLO) is a popular neural network-based real-time object recognition algorithm used for identifying objects in video streams. However, YOLO was originally developed to work in environments where computing power is not a matter of concern. The objective of this study is to explore the possibilities of implementing the original YOLO or its variants in an embedded setting where computer power and resources are limited. Potential adjustments to the YOLO configuration will be proposed and the effects studied. Appropriated embedded computer hardware will be tested. Studies will also be conducted to address issues related to actuators and motion control algorithms. The result of this research will facilitate the application of YOLO in robotic tasks where good computer vision is required.