A Dynamic Disparity Range Aggregation Method for Multi-baseline Stereo Matching

Depth estimation is a pivotal challenge in the realm of signal processing, finding various applications in fields like robotics and autonomous systems. Multiple cameras are used in these applications and are found to be very useful. In this paper we address the problem of obtaining the depth information from images with improved compute complexity and accuracy. The proposed algorithm consists of three major steps, namely (a) Initial cost volume calculation, (b) Iterative calculation of successive cost volumes and (c) Aggregation of cost volumes. We use a fusion of simple cost volumes to get the initial disparity map. To improve compute complexity, we propose a novel algorithm which reduces the search range and functions as an extension tailored to overcome the limitations of the Trinocular Dynamic Disparity Range (TDDR) algorithm. Results are shown to demonstrate the performance of the algorithm with a 61.71% decrease in the computational time, compared with an existing method, on multiple Middlebury Stereo datasets.