SLAM for drone

ReemOthman

Hello,
My application is to integrate a drone with SLAM for indoor inspections.
Would you suggest me which of your cameras will be the best fit for our application to make a complete point cloud with full coverage?

and would be useful if I use OAK-FFC-6P board for connecting 6 cameras (3 upward & 3 downward)?

Finally, can I use the 6 cameras with jetson nano board without using the OAK-FFC-6P board?

Thanks in advance.

erik

Hi
Yes that would be possible, you can run sparse SLAM on the quadcore arm on-device (for RVC3), example here:

But you can also run SLAM on host computer (like jetson) and use OAK (RVC2) just as a sensor, some examples here: https://docs.luxonis.com/en/latest/pages/slam_oak/

Yes, that would work, just note that stereo hardware resources would then be split across, which means you wouldn't be able to achieve such a high FPS. Yep, all OAK cameras work without a problem with Jetson devices.

For camera sensors, I'd suggest OV9282, as they are most widely used/supported (and have best performance).

Thoughts?

ReemOthman

This video using RVC3 OAK-FFC-6P board or what?

and you mean I can use RVC2 for SLAM, but it says here just RVC3 ?
Or am I confused?

l4es

Hi ReemOthman,

I think that the doc means RVC2 has not supported on-board VIO and SLAM yet. And on-board means that the VIO and SLAM would run directly on the RVC2 without any additional platform (such as Jetson or Raspberry Pi 4 or x86 laptop). But this is not the case of RVC2 but RVC3.

Meanwhile, Erik meant that you could use RVC2 for SLAM but for this case the RVC2 could only be the sensor (RGB + depth + IMU) and the SLAM would run on the host which could be the Jetson or the Raspberry Pi 4 or the x86 laptop taking the input data from the aforementioned sensor.

ReemOthman

Thanks a lot @erik @l4es for your clarification!

I have another question, since I can use the OAK-FFC OV9282 (pair) cameras as depth camera with Jetson without the OAK-FFC boards, what is the exact functions of these boards?? I mean is there is any feature I'm gonna lose If I didn't use these boards Or Jetson will do all their function ?

l4es

Hi ReemOthman ,

I would say that the OAK-FFC OV9282 is just a MIPI-CSI2 based camera/sensor module. You can use it with other platforms disposing the MIPI-CSI2 Rx interfaces (with or without the need of an adapter board so that the signals match one by one).

Otherwise, OAK FFC boards have Intel Myriad X (RVC2) or Keembay (RVC3) VPUs that provide not only vision functionalities such as ISP, H264/265 encoding, but also depth mapping and deep learning which are usually heavily computational for platforms without hardware acceleration (such as CUDA in Jetson).

Currently, with firmware in OAK FFC boards, your selected OAK-FFC OV9282 will work out-of-the box. You will not need to develop the kernel driver for it or to do the ISP calibration / tuning which could take a lot of effort and time (based on my experience with Nvidia Jetson and other platforms such as NXP or TI).

Also, RVC3 based OAK-FFC boards = RCV2 based OAK-FFC boards + Raspberry Pi 3/4 (not 100% exact in terms of CPU cores).

ReemOthman

Aha okay now I understood.

But do you have a personal experience using Oak cameras for depth vision with jetson without the Oak boards?

l4es

Hi ReemOthman,

As a low-level system integrator, I have experience in developing the kernel driver forthe camera modules similar to OAK FFC cameras (AR0234 and others) with Nvidia Jetsons, but not with the depth vision on them yet.

ReemOthman

Understand …Thanks a lot @l4es for your help!