I'm having the same trouble, even though I've taken 50+ photos of the same person standing still, Photoscan tries to "align" all of my shots into a clump at the front of the person. As you can guess, the "matching points" cloud is mostly noise, and I don't even get close to anything about model generation.
I've been trying to make Photoscan work for a while now, and I keep getting the same problem.
My model is my wife, who would prefer not to have her pictures on the internet, so I won't post screenshots here, but maybe my workflow is wrong. I put masking tape on the floor in a circle, and then mark 16 points equidistant around that circle. I have my wife or whoever I'm trying to photograph in the center, and then I take shots at each point around the circle, at four different heights - my eye level (6ft), crouching, squatting, and then right against the floor. Each shot overlaps the ones around it, and each shot is taken "flat" (i.e. not tilted, as much as possible) towards the object. I get about one third to one half of the person I'm trying to shoot with each shot; i.e. shot 1a is front, head and upper torso; shot 1b is front, torso and waist, shot 1c is waist and thighs, shot 1d is thighs and feet. Shot 2a is 22.5 degrees counterclockwise of front, head and upper torso, and so on. (They're not named that, though - the images are all just named image001.jpg, 002.jpg. etc., if that makes a difference.)
I tried loading this selection of shots into Photoscan, but again, the alignment woes I mentioned before. After reading this thread, I tried masking the photos very carefully so that only my wife was in the shots, with the same results - random noise of (very few) points, and the cameras "aligned" in a clump in front of, not around, her.
I'm trying this out with a dinky little point and shoot (12 MP), so that may be the problem; with indoor lighting (which is, unfortunately, the lighting I will have to use if I can get this to work for my application.)
I'm going to attempt it with a DSLR later on today, and hopefully that will fix the problem. Otherwise, am I missing anything important? Is there any thing that I am screwing up, theory-wise? (Clearly, I'm screwing up something in practice.)
Thanks so much for any help; it's greatly appreciated.