If you're of the mindset that Kinect on the Xbox One is simply a more advanced version of what we saw on the Xbox 360, you'd be grossly underestimating it. In a TechNet blog post, Microsoft goes into some great detail to explain what makes its next Kinect so amazing - and it's not just one or two things; it's a variety of things. I'll be honest in saying that prior to reading through this and checking out the videos, I had no special respect for Kinect - but my opinions have pulled a 180.
As the video below shows, Kinect can read people three different ways. For starters, it can convert the person standing in front of Kinect into a skeleton with numerous joints, where even the closing of a hand will be represented. In the video, the person moving around in front of the Kinect is perfectly reenacted by this skeleton. Further, there's a "Block Man" model as Microsoft dubbed it, where Kinect can understand the orientation of your body parts - turning your leg inward, for example, wouldn't be oblivious to Kinect.
Also in the above video is a "Muscle+Force" model which shows you in real-time how much force you're applying to any part of your body. In normal standing mode, you'll be applying a little force to your legs, while if you raise one leg up, you'll be applying a lot more force to the other one. As expected, all of this is perfectly understood by Kinect. Finally, there's also a way for the sensor to detect your heart rate based on your face alone - clearly, the potential there is huge for fitness titles.
A second video, seen below, explains how well Kinect can gather data on someone standing in front of it regardless of the lighting. In effect, the camera will turn all objects in the near scene into a 3D object, including of course the person standing in front of it. With all of the technologies mentioned above, Kinect can easily understand what you're trying to do. In a dimly-lit room, Kinect will switch to IR mode, which renders any lightsource invisible. Even in this mode, Kinect's tracking abilities are accurate.
Helping to make all of this possible, besides the software of course, is the the "time-of-flight" camera, which emits light signals to effectively build a scene in its mind in realtime. These cameras need to be accurate to 1/10,000,000,000 of a second - this is speed of light stuff. Mind-boggling, to say the least.
Dear Microsoft: It wouldn't have killed you to release all of this information long ago. This is genuinely impressive stuff.