Move frame comparison technique ops to GPU

[lyuzinmaxim]

In #88, we created the pipeline which demostrates the modular approach to skip the frames based on the comparison of consecutive ones. In this ticket, we just need to move frame comparison operations to GPU, and make sure that we don't do the GPU->CPU->GPU copies.

Please use OpenCV+CUDA instead of cupy.

Links:

• How to access raw GPU frames (directly, w/o copying to CPU): https://docs.nvidia.com/metropolis/deepstream/6.4/dev-guide/text/DS_sample_custom_gstream.html#accessing-nvbufsurface-memory-in-opencv
• The only way to access the GPU surface (directly) on Jetson devices: https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/blob/9b27f02ffea46a3ded2ad26b3eea27ef3e2dfded/bindings/docstrings/functionsdoc.h#L571

DoD:

• a new script is created (based on CPU script)
• frame comparison is working in the same way on GPU as on CPU

[lyuzinmaxim]

Please do the following as a part of your ticket:

• change src/frame_change_detector.py to use the same operations/algorithms as you use in your GPU-based version
• make sure that both CPU and GPU versions are producing exactly same results. If not, please raise it.

[lyuzinmaxim]

scp

scp video_on_windows_local_machine.mp4 <USER_NAME>@<SERVER_IP>:/home/<USER_NAME>/...