Well that might work.
If it's just comparing the face in front of the camera against an earlier one on record then it might do a reasonable job. The problem comes when trying to compare a random face against a large dataset. Then you will run into the problem of trying to keep false positives at a reasonable level without introducing false negatives. The problem is still not close to being solved for large datasets (e.g. a country and all previous visitors.) Indeed, humans would have a problem even if we could remember 70 millions peoples' faces. Some people just look quite alike. It isn't much of a problem for us at the minute as we only remember a few thousand faces but if we could scale our memory up we'd run into the same problem.
Some companies were trying to flog this stuff at a Retail expo the other day and it worked a treat but that was with only a small number of faces.