numDisparity determines the resolution of your stereo map. If you define, for example, minDisparity of 20, the algorithm will attempt to define 20 different levels of depth for your map The higher this value, the higher resolution you get for you depth map, but the algorithm also slows down and will return less coherent results for a very demanding parameter.
minDisparity is usually zero, it means the least distance the algorithm is looking for.
BlockSize is the size of the block of pixels the algorithm will compare for each stereo pair.
perhaps you should share your rectified image & your parameters!