A comparison of cost construction methods onto a C6678 platform for stereo matching
Judicaël Menant, Guillaume Gautier, Jean-François Nezan, Muriel Pressigout, Luce Morin · 2016
Stereo matching techniques aim at reconstructing depth information from a pair of images. The use of stereo matching algorithms in embedded systems is very challenging due to the complexity of state-of-the-art algorithms. Most stereo matching algorithms are made of three different parts; the cost construction, the cost aggregation and the disparity selection. This paper focuses on comparing the efficiency of different cost construction methods implemented on a Digital Signal Processor (DSP) C6678 platform. Three cost construction algorithms based on census, Sum of Absolute Differences (SAD) and Mutual Information (MI) have been compared in terms of output quality and execution time. Each method has its own specificity discussed in this paper. The SAD is the simplest one and is used as reference in this paper. The census has a good output quality, and the MI is faster.