0

I have two stereo pairs with different baselines. One with 150 mm and the other with 500 mm. I need to find out the working range (in meters) for both the setups.

I use triangulation to generate 3D points from disparity maps. I know already that using a shorter baseline setup provides better depth accuracy in near range and the longer baseline setup provides better depth accuracy in the far range.

I need to find the distance value after which depth accuracy with shorter baseline is no longer trustful when compared with longer baseline setup.

Thanks in advance :)

Beginner_In_Image_Processing
  • 3
  • 4

1 Answers1

0

You are not entirely correct. Bigger distance between cameras is always better for accuracy in whole range (not just for far objects). The problem with near objects and big camera distance is that the FOV overlapped area of both cameras is smaller then if the cameras are closer together so you have less space to work with.

To determine how far the cameras should be you need to define what you want. I assume you have some far distance accuracy you want to achieve. So compute Z-step in that distance with Angle step defined from the resolution of the cameras. From the triangulation equation derive the camera distance so the Z-step is in you accuracy ... Then check the near object are actually in the FOV of both cameras and if not compromise between far accuracy and range.

Spektre
  • 49,595
  • 11
  • 110
  • 380
  • Thanks a lot for the explanation. However, I am not trying to determine the best baseline distance with a preset depth accuracy. I have two stereo setups as mentioned before. For example, a person is standing at a distance of 10 meters (ground truth). The shorter baseline stereo calculates the distance of the person say 10.2 m and the longer baseline approximates it as 10.6 m. Clearly in this case the measurement from shorter baseline camera is more closer to the actual distance. – Beginner_In_Image_Processing Nov 18 '16 at 10:27
  • Now suppose the person moves to a distance of 30 m. Now the longer baseline stereo measures 30.3 m and the shorter one measures 30.8 m. In this case the longer baseline setup provides a more accurate measurement. My question to you is, at which distance, the values of longer baseline stereo start to become more accurate when compared with the shorter baseline setup. @Spektre – Beginner_In_Image_Processing Nov 18 '16 at 10:31
  • @Beginner_In_Image_Processing there must be something fishy about your setup ... do booth setups have the same resolution? Is standard triangulation used for depth perception or a different method? or you just have wrongly calibrated one or both setups, It also can be due to aliasing causing wrong feature coordinate detection ... and may not be the case for different images) – Spektre Nov 18 '16 at 15:59
  • @ Spektre No no, I think you misunderstood, Both camera setups give quite accurate depth readings, it's just that to reduce the number of 3D points, I have to define regions. For example i select all 3D points between 5-15 m with the shorter baseline setup and above that I use 3D points from the larger baseline setup. – Beginner_In_Image_Processing Nov 18 '16 at 16:05
  • The idea is to divide the scene into two regions near and far. And for the near region I need to use 3D points generated with the shorter baseline setup and for the far region with longer baseline. – Beginner_In_Image_Processing Nov 18 '16 at 16:10