Yes it is possible, but you need to extend the setup a bit. I have recently created a library to run ML algorithms for various kind of image detection tasks. One network can be used for hand detection, which then can be used to extract the image of the hand. You can of course use your kinect, but for this you will need the extracted image of the hand.
This image then can be fed into an image classifier, which then tries to recognise the gesture (thumbs up and so on). Of course it is also possible to detect patterns in movement (check my second answer).
I have created a simple example which uses my library, together with wekinator to classify hand images. The region of interest of the hand is converted to a binary image, where all the skin-tone (only wester & asian skin tones atm) parts get white and everything else black. This is done by extracting the hue of the skin tone you would like to track.
Then the image will be resized to 16x16 pixels and sent to wekinator. Wekinator is then trained to seperate those 416 pixels into two classes: thumb up & normal.
Check out the example here:
HandGestureRecognition
But you may have to install my library (deep vision) first and of course wekinator. The wekinator project is in the example folder too.