This problem result is a Cloud/Mesh looks like a messy cloud of noise and not a surface.
This can result when:
Orientation for that pair is not good enough (residuals too big). Note the lower the base-to-height ratio the better the orientation needs to be.
Solution: Strive for a maximum residual under about 2 pixels. If the photos being used (for the Paired Photos method) have a very small base-to-height ratio (less than 0.1) then strive for less than 1.0 pixel maximum residual. Study your points carefully and adjust or retake photos using targets for the orientation points.
The depth range setting is too small and does not match the reality of the distance of the real surface from the approximating starting Surface.
Solution: Applies to Paired Photos only. Under the Basic DSM settings there is Depth Range. The program sets a default range but it is a guess based on either the 3D points in the project or the selected surface. If the range is too small no good matches will be found within the search distance or bad matches will be found. The stronger the project (lower residuals, non-moving object/cameras, and strong random texture) the larger this range can be and you will still get good results. In a strong project the range can be far larger than it needs to be and good results can still be obtained. Shrinking the range is needed for projects that are not as strong.
Base-to-height ratio is too large for one or more photo pairs.
Solution: As the camera positions get too far apart (and usually angled more to get the same part of the object imaged in the pair) the surface starts to look too different in the images. The matching cannot be done successfully with too large a difference. For Paired Photos, choose pairs of photos with base-to-height ratios under 1.0 (under 0.5 better). A base-to-height ratio down to 0.05 can be processed but only for very strong high-quality projects. For MVS ensure there are photos with smaller angles.
The sampling density is too low.
Solution:
• MVS: Decrease the Surface density, decrease Resolution level, decrease Minimum angle (but note that more noise may result).
• Paired Photos: In some circumstances when the sampling density is too low the optimization algorithm picks the wrong matches for the image samples. The number of potential matches per row needs to be increased so there is a greater probability that the correct match (and hence the correct final 3D point) is chosen. In the DSM Options dialog try making the Sampling Density value smaller (try a factor of 2 to 10 smaller at first). The DSM processing will take longer and the number of points returned will be higher but the amount of noise may decrease. If the resulting density of points is too high for your application you can use the Modify Mesh tools to decimate the cloud.
The approximating base surface is too far away from the true surface.
Solution: Applies to Paired Photos only. For the search range to be reasonable the approximating base surface (i.e. defined implicitly by the 3D points in the project, or by a selected surface) needs to be close to the real surface. Some projects will work with the base surface not being exact but it depends on the randomness and how visible the texture is. If you are not seeing any data or very little this is an important item to check. Make sure the base surface is closer to the real surface.
The object moved between photographs.
Solution: Applies to both MVS and Paired Photos. The object needs to be very steady between photo pairs being processed. It should be completely stationary or move less than 2 pixels. If you are modeling an object you know will move then multiple synchronized cameras must be used.
A flash was used or some lighting source moved between photographs causing shadows and textures to change.
Solution: Applies to MVS and Paired Photos. If you can see shadows (large or small) move between photographs then this will cause mismatches and noise. Do not use a flash mounted to the camera if the camera is moving. Light sources that do not move between photographs are best. Some Lambertian smooth surfaces can be modeled with a small amount of light source movement.