Can you please add the OpenCV library and LUACV to the installation?
I want to use it to utilize the camera and some image recognition for a sanity check of the homing, especially when picking up the tools.
I use OpenCV on a secondary system and secondary camera(s) and just send movement commands to the farmbot ecosystem. I can imagine the overhead of OpenCV on some of the RaspberryPis while running Farmbot may be a bit much especially with multi-streams and rtsp work being as taxing. I am having fun with an Nvidia AGX unit mounted on a secondary box on the opposite tower of the gantry which send commands to the farmbot OS remotely. I went with Mosquitto MQTT for messaging. The Nvidia supports upto 16 GMSL, I recommend Globalshutter ones vs the Borer camera that comes with the farmbot especially seeing your plants during movement
2 Likes
1 Like
My new arrangement in progress.
I used a second 24v power supply to feed an older farmbot electronics box with two terminal blocks, then used a buck voltage regulator to get 19.2v into an Nvidia AGX system, a motorcycle cigarette lighter style with inline fuse to usb port for 5v to a Wi-Fi camera. I also put a 24v led light on the terminal to quickly show powered up and ready to go. I did add antennas to the AGX , one trick is if you get them snapped in to the micro adapter, use some hot glue to keep them in place,
Next steps wire up the GPIO ports for servos and relay switches, the servos control the 3d cameras on a custom tool mount. Both farmbot commands , opencv and tensor flow is handled at the edge.
2 Likes
@jturbett This is great work! It would be really interesting to see video of the tensorflow output from your system.
What is your opinion of the Nvidia AGX Performance in this application? How do you think it would compare to the Raspberry Pi 4B -16 GB RAM ?
The AGX is a different animal and remember I use it with the farmbot not in place. It is a purpose built for AI workflow processing. To think my first super computer in school was the size of a Walmart store in the 80’s and now this is more powerful compute at 35 watts in the size of my hand… (I won’t debate inference latency metrics here, but A GPU accelerator at the edge is a powerful tool) also a jetson nano might fit well on a farmbot but mine is currently roaming the yard learning to avoid wolfhound poo…) the Nvidia line offers more flexibility than a Coral edge TPU to have other workloads concurrent , this AGX also can scale the GPUs to the power level and performance that can be rationed vs the TPU could melt plastic nearly idle. The high speed GSML cameras make a real difference( maybe not on plants) but in other applications I work on.
So to compare to a Raspberry Pi for price , performance , watt to compute and ease of use, well the Nvidia loses, add concurrent workloads, opencv (with CUDA ) , inference without an accelerator, I’ll show you the drawer of dead raspberry pi’s that me and my staff fried. ROS w raspberry pi can be frustrating and disappointing. A Raspberry 4B opencv with a high speed usb camera doing LPR (license plate recognition) means every 3rd car gets recognized 
So apples and oranges. Or in my case ripe or not ripe peppers. Which got planted by seed this past weekend, and with me flying this year, my challenging water line stays together the system will watch them grow.
2 Likes
Oh on the AI computing, Let me see if I can use a video I did for my team.
2 Likes
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.
Are you able to share the video of the tensorflow output from your system?
You mentioned that you might be able to share a video you did for your team.
1 Like