Exploring the Art of Visual Motion Estimation in Human and Machine Vision Systems

Visual motion estimation includes the ability to infer the speed and direction of objects/elements in a scene based on the dynamic inputs from a visual sensor like a camera or eye. Such a function is crucial for humans and animals to navigate and interact with the dynamic world, and it is equally vital in the engineering field for machine/AI -agents to process downstream visual tasks, like object tracking, video interpolation, state prediction, etc. As such, visual motion estimation has garnered attention in both human science and engineering fields, such as human visual motion perception in vision science (VS) societies and optical/scene flow estimation in computer vision (CV) This thesis examines the mechanisms of visual motion estimation at the intersection between the CV and VS. The final goal is minimizing the Gap between CV and human vision systems by constructing a generalized motion processing model