Compare point cloud with CAD model, for object recognition?

edit

I am trying to do industrial part recognition using my robot. I am using a Kinect for as my vision component for the robot. The kinect fetches me real clustered segmented data ( i wrote a code for that). Now i need to perform recognition using a given CAD model. I thought converting it into .pcd file would be one option however, I dont know how feasible and accurate final output I will have after converting a .assembly --> .STL/Mesh--> .PCD. Can some one tell me how can the comparison done to perform the recognition. I am using ROS indigo, PCL (latest), and Kinect Xbox 360. Also all this needs to be done in real time and not offline. As detection and segmentation has been achieved already in real time.