# What is scale ambiguity in computer vision?

When you compute relative pose between cameras, translation is only accurate upto a scale. Need some help in understanding the scale ambiguity and what exactly does it mean translation is accurate only upto a scale.

Consider a case where you have 3 camera positions C1, C2, C3 at vertices of right angled triangle with 5,12,13 meters(C1C2, C2C3, C1C3 lengths in meters) as sides of triangle. Assume that you have unlimited correspondences between images I1, I2, I3 captured by camera at C1, C2, C3 and you can accurately measure pose.

Now when you compute pose of C1, C2 using correspondences between images, let translation be x. Similarly let translation between C2,C3 be y.

1. is x/y always 5/12 like the case where scale is known 5meters/12 ?meters
2. Since scale is not exactly known, can x/y be 5meters/12kilometers(5m/12*1000m = 5/12000) ? Is this this what scale ambiguity means? In other words does it mean that C1C2 translation could be in meters, C2C3 could be in kilometers.
3. Can this not be resolved using 3rd camera and by using Vector Addition. C1C2+C2C3 = C1C3? There by ensuring x/y is always 5/12? with help of 3rd camera.