Can binocular vision produce accurate point clouds? It seems acceptable for measuring the distance of a single object, but are the point clouds reliable?
The prerequisite for using binocular disparity is to know how the objects in the two images are matched. Which object in the right-eye image corresponds to an object in the left-eye image.
This can be automatically calculated using feature extraction algorithms such as SIFT, combined with block matching algorithms like SAD (Sum of Absolute Differences) and SSD (Sum of Squared Differences), without the need for manual annotation.