1P1-V01 Coarse-to-fine法を取り入れた生体模倣型実時間両眼視差検出システム(ロボットビジョン)

Abstract

We developed an algorithm and architecture for calculating the optimal binocular disparity efficiently based on the binocular disparity energy model with a coarse-to-fine strategy. Because the model requires a lot of spatial filtering whose computational cost is very high, we designed hardware architecture so that it executes the required spatial computation in real time by using a resistive network and a field programmable gate array (FPGA). The resistive network enables fast multiple band pass filtering required for implementing the disparity model. The algorithm takes advantages of the hardware architecture to calculate the optimal disparity from multiple band-passed images. The optimal disparities calculated from band-passed images with a low and high frequency offer a wide detectable disparity range and high precision, respectively. We evaluated the algorithm by using a software simulation. The results showed that the algorithm offers both a wide disparity range and high precision.