Hardware selection recommendations for a embedded like OpenCV project

Question

I need to choose hardware for an OpenCV project that I am working on currently. Space is limited hence I cannot use a full blown computer. The project specs are listed below:

1. Using two USB 3.0 high speed Basler cameras running at 300 fps each (300 megs/sec each.) The video shot is between 1 to 1.5 seconds at a time.
2. I am using the LBP (local binary patterns) OpenCV library to detect objects in the video.
3. The video processing should be real-time or as close to real-time as possible - a lag of up to 2 seconds between the video end and result extraction is acceptable.
4. The cameras are used for two different tasks, obtaining the angle of an object moving in parallel to a camera and its proximity to the cameras - hence the need for stereoscopic vision.
5. To minimize the data quantity, basler offers me the ability to skip frames, which I have done. My project works if I only use every 4th frame. Hence I only need to process 225 frames at most. More optimization is possible by cutting the duration of the video to below 1 second.
6. The hardware needs to support Linux - my OS of choice
7. Ideally I would like the OS loaded from an SD card
8. The hardware needs to have 2 USB 3.0 slots
9. HDMI or VGA output is needed.
10. The hardware should not be any bigger than 8" x 8" x 3"
11. Internet via (RJ-45)
12. Price should be around $500 - $1000
13. Memory - ideally at least 4GB
14. CUDA hardware acceleration or similar would be very nice
15. Power - no restriction - city grid
16. Needs to interface with a SSD drive as well

So far I have looked at an Intel NUC and a Nvidia JetsonTX2. Any other suggestions are greatly appreciated.