It's not sufficient for the photos just to cover the area; every area needs to be covered by a bare minimum of 3 photos whose orientations are not so different that the appearance of the surface in each image can not be matched.
Also, if your images are slightly noisy, blurry or out of focus then dense cloud reconstruction might fail in some areas, but may be improved by processing at a lower quality level. I.e. if you process the dense cloud at highest, but the images at pixel level are not representative of the surface (due to blur/noise), then dropping back to high or medium which only looks at every 4th or 16th pixel may help fill some of the gaps as this effectively sharpens the image (albeit at a lower resolution) and means lower chance of picking up an outlying noisy pixel.