In fact, hand posture is a collection of pictures of hand (image sequences or video data). One picture - frame. From frame, we can detect hand. From multiple frames, we can predict the hand posture. So, you only need one camera. If you want to track the hand posture in 3D space, you need two camera or use special camera, just like true depth camera.

As you stated, one single IR camera can provide you the distances. Multiple IR can introduce you interference due to the light of each camera. You can use a IR camera in combination with a color camera just like the Kinect works.

We use all of our 9-camera Qualisys Oqus cameras to track the head, trunk, and upper limbs: moving an object in the hand from the lateral side of the hip to a position in front of the participant. Anatomical markers can be placed at the ulnar and radial aspects of the wrist, and on the dorsum of the hand, and digits, so the palm is free for grasp. Each marker must be seen by at least 2 cameras to have x, y, z coordinates in the calibrated volume. Even if the hand remains palm-down (forearm pronated), the minimum number of cameras would be 2. However, increased range of hand positions and postures may cause markers on one side of the hand to be obscured from view by another part of the hand, requiring at least 3 anterior/overhead cameras (e.g.: left, right, central) As the number of hand segments represented and hence number of markers goes up, marker diameter may have to be smaller, and cameras may have to be closer, and greater in number. 5 of our cameras (central, left-high, left-low, right-high, right-low) are all important to track the hands during this reach and grasp sort of task, in which the forearm may be pronated, supinated, or neutral, and the digits may be flexed, extended, or in opposition. I echo Marco's addition of a colour camera (especially if you can project the 3D marker coordinates onto the video's 2D plane), to provide a familiar movie image, and a means of checking marker identification tracking and subsequent angle calculations.

In fact, hand posture is a collection of pictures of hand (image sequences or video data). One picture - frame. From frame, we can detect hand. From multiple frames, we can predict the hand posture. So, you only need one camera. If you want to track the hand posture in 3D space, you need two camera or use special camera, just like true depth camera.