Light field data processor design for depth estimation using confidence-assisted disparities

In this paper, we propose a hardware-compatible light field data processing algorithm for depth estimation. This method combines Average Absolute Difference (AAD) technique and a confidence weighted check to calculate disparities between sub-images, from which depths of objects in the scenes can be obtained. We implement the algorithm using both software and hardware approaches. In the hardware design, we reuse the circuit block for 0/90-degree AAD comparison in the first pass, and 45-degree AAD comparison in the second pass to balance hardware complexity and computing speed. The speed-up of more than 12 times can be achieved in the pipelined version.