Stereo Disparity Calculation Based on Superpixel Clustering for Assisted Driving

The calculation of disparity in stereoscopic registration is a critical step in performing the 3D reconstruction of an environment. Typically, disparity calculation is performed on a pixel-by-pixel basis for one of the images in the stereo pair. However, this paper proposes a novel algorithm that calculates disparity using superpixel clustering. The proposed method begins by segmenting one of the stereo images into superpixels, which are then used as comparison windows in the Census Transform algorithm, replacing the traditional square or rectangular windows. Additionally, the algorithm incorporates the color difference between windows, enabling a comparison that accounts for spatial, textural, and color differences. These enhancements make the algorithm robust to variations in illumination and low-texture regions. Furthermore, the use of superpixels significantly reduces the computational complexity of the process. By decreasing the number of operations required to compute the disparity map for each stereo pair, the proposed method achieves improved processing speed while maintaining a low error rate.