@Chuck 's comment is correct; You can have some sort of polling software which detects some image (continuously tries to detect an object from image stream).
Else, if this does not work, then, it is better to add some other sensor like proximity sensor or so to detect objects like human and then turn on thermal camera if a threshold is crossed. But, as @Chuck mentions, unless the objectives are known, it is difficult to help you out.
If an image-array is obtained, then image-processing can be done through many libraries. Some of the libraries which I remember are OpenCV, MATLAB, PIL (Python Imaging Library), BoofCV, JavaCV, SimpleCV etc (depending on your programming language).