Analog VLSI Circuits for the Perception of Visual Motion

8

Comparison to Human Motion Vision

This book began with a description of the demanding task of a football goal-keeper estimating the relative motion between the ball and himself. Yet, aside from this initial illustration of the problem, the development of the optical flow network architecture, its network extensions for motion segmentation and selective attention, and their aVLSI implementations were purely guided by a rational perspective on what properties a visual motion estimation system has to have in order to solve the task. At no moment was there a motivation to imitate any existing biological solution as was the goal in other approaches.¹ Considering the large differences in the computational substrate between silicon and nervous tissue, a close structural embodiment of biological solutions does not seem necessarily reasonable.

However, the rationale and logic that guided the development of the presented networks is human, and one might wonder how far they are influenced by the human perspective on the visual world, that is by the human visual motion system. This chapter will draw a comparison between the presented silicon visual motion systems (specifically the smooth optical flow chip) and the human visual motion system. It will compare the perceptual behavior of human subjects and the chip in some typical psychophysical experiments and loosely discuss similarities in the computational structure of both systems.

8.1 Human vs. Chip Perception

The human visual motion system ...