QUANTITATIVE COMPARISON BETWEEN NEURAL NETWORK- AND SGM-BASED STEREO MATCHING
Keywords: Stereo Matching, Deep Learning, ZED Camera
Abstract. Over the last decades, various methods for three-dimensional detection of the environment have been developed and successfully used. This work considers classical stereo methods, which can determine depth information by the means of correspondence analysis on the basis of two pictures of a scene. Recently, neural networks have been used to solve correspondence analysis. These procedures came first places on corresponding benchmarks and are ahead of many already established solutions. In this work, images captured by the ZED camera are evaluated for accuracy of the depth maps generated by several approaches. This includes modern methods based on neural networks.