The straight binary encoding wouldn't be able to position LED number zero, since it would be turned off in all pictures. And I think the algorithm will probably be more robust with LEDs that are turned on in many pictures compared to those which are only turned on in some of the pictures. You can restrict it to assign each LED a number in which exactly half of the bits are one and half are zero. It will still be logarithmic time, but with a slightly higher constant factor. An LED which is not visible from both camera angles will obviously be a problem. That problem is caused by the positioning of the cameras and will be the same regardless of whether you use the logarithmic or the linear algorithm. If you are encountering that problem, I think it's best addressed by using more than two camera angles.