This is normal.
The three processes are
- NI mate main process
- Kinect for Windows v2 sensor program. This handles the communication to the real hardware and gives the data to NI mate main process
- VTDecoderXPCService: A hardware accelerated JPEG decoder.
The Kinect for Windows v2 outputs the color feed at full HD with JPEG compression. NI mate uses libfreenect2 for handling the sensor, and libfreenect2 has several options for decompressing the JPEG image. In this case, it's using the VideoToolbox API which spawns that high CPU usage process. Apparently this is not using hardware acceleration yet though, now that I read about it.
The JPEG decoding is largely "fast enough" in libfreenect2 right now, but hopefully there can be better hardware acceleration. Some of the performance research can be read at: https://github.com/OpenKinect/libfreenect2/wiki/Performance