2 of 2 people found this helpful
The FPGA and its carrier board are pretty amazing, but I suspect that unless anyone has already experience with the technology, it may require quite a bit of effort to learn how to properly use it.
2 of 2 people found this helpful
Not just Zynq experience, but extensive video processing on Zynq experience. I've some road test experience with Zynq and dabbled in the video side with the Zybo Z7, but definitely couldn't do this board justice. I suspect finding anyone to adequately road test this will be very tricky.
Looks like an interesting board. I'd like to design a real-time PIV system someday, but no time right now.
So many projects........
I love the project. I tear up not having the experience to undertake such a project.
I have designed FMC daughter cards for Xilinx boards, and would enjoy embracing this opportunity to road test the UltraZed-EV.
Impressive, but unfortunately I have other priorities at the moment.
I think, I would be interested in something like this too...
As previous experience, I have these:
- Stereo Vision and LiDAR Powered Donkey Car - hardware accelerated stereo vision implemented on an Ultra96 board (based on Xilinx UltraScale+ ZU3EG)
- PYNQ-Z2 Dev Board: Python Productivity for Zynq® - Review - road test of the PYNQ-Z2 board (based on a Zynq-7000 series SoC)
As a Road Test project, probably I would try some object detection using hardware accelerated BNN-s (Binary Neural Networks). The video input / outputs could either HDMI or network video streams (RTSP).
I would be interested in testing this board.It looks feature packed.
I would port PYNQ to this board if it's not already. I ported PYNQ to the ZC702:
I have posted a couple of similar projects:without AI.
Ideally I would use two video feeds to implement an image captioning neural network (CNN+ LSTM). I see the carrier has HDMI in which is great.
I have an idea on how to do this on a SOC but I will probably need to use PYNQ framework to speed up development.
This will make testing IP cores quite straightforward since I one would have to use Python to talk to them.
Then I would implement a video mixer with multiple cameras. Will use HLS IP's in order to meet the free cores requirements.
The MIPI-CSI core is quite expensive. Word on the vine is that Xilinx plans to release it for free in the future Vivado versions.
so I would use either an HDMi video source, a USB camera which will necessitate Petalinux or use a custom FMC solution with multiple cameras.
As for AI I think one can either run the DPU V3 .One other option is to compile and use YoloV3 for object localization running on the PS
- I am interested in road testing this board since I can provide a very unique and comprehensive review on this board, the reason being that I am working on thermal and hyperspectral imaging. Most of the other reviewers will never deviate from visible imaging due to its very different handling approaches and algorithms.
- My project has been supported by various industries and companies, who want my research and application to reach a conclusive end. Hence, I can submit a detailed, comprehensive and extensive road test, which will have an on-site demonstration in real-world environments.
- My research, if successful will significantly impact agriculture in India and even worldwide. I am looking to create a huge dataset, using various thermal cameras (AMG, MLX, FLIR, etc) and use CV/ML/AI and even NN (I am experimenting with various architectures and methods to get the best result) and then train the model on a portable SoM specifically meant for video processing. I have tried various devices and systems already, like the Jetson Nano but they are falling quite short of meeting my requirements in the budget I want to deliver the product at.
- My project will also contain a UAV drone, which is currently being designed and will make proper use of video capture and processing features of this board. External pieces are COTS and will have optimum VCU utilization.
As far as my background is concerned, I have just recently joined the roadtest community, bagged my first road test as soon as I joined a few days back... My team has represented my country Internationally, in the AAS Cansat competition, sponsored by NASA securing a worldwide ranking of 34 last year. I have won National hackathons and competitions. Apart from all this, I have bagged hardware on hackster.io multiple times for my different ideas. I can provide any evidence/ photos for these projects and competitions if needed.
I don't want to detract from your application, but I had a question:
For thermal imaging, and OpenCV, isn't even a low-end processor adequate? I've used a Flir chip with a BeagleBone Black (it is almost a decade old tech) and that was fast enough to run OpenCV performing some operations, at the maximum frames-per-sec I could pull out of the chip. The other parts you refer to (Panasonic, Melexis) wouldn't begin to warm up first-gen Pi. Using a combined FPGA+processor is quite different..
- I have tried various devices and systems already, like the Jetson Nano but they are falling quite short of meeting my requirements in the budget I want to deliver the product at.
You do know the price difference between the Jetson Nano and this Xilinx part, right?
Yes openCV etc can be used on a simple controller. But all applications differ in size and requirements. OpenCV is something that is not sufficient for some of my applications. I want to develop my system using MATLAB, since I find it really easy to interface it with the rest of my system, and even simulate the environment i intend to deploy my solution in. It helps me test my multi domain systems and projects at one place.
The parts that I have mentioned have been selected as per my current hardware. If I get a chance, I will buy better and more powerful peripherals that will serve my purpose better. The modules that I am currently using are good for prototyping, but not for a real world scenario I feel. I am very well aware of the beaglebone, rpi, etc. They are good for playing around, not for actual deployment. I want to move past that, all I was suggesting is that I am already working on those things. Thats all.
Also, as I mentioned, I have worked on a multitude of projects, not just that one. There are other suitable projects that I am working on as well and I feel I will be able to write a good and extensive review.
Lastly, yes I am aware of the pricing of this board. It won't be a part of the deployable solution, I intend to use it more for R and D and applications where I need a more powerful hardware. I have been considering the Jetson Xavier as well, lets see.
I hope I have answered all your doubts.
3 of 3 people found this helpful
No, I still have doubts.
On the one hand you say:
"The modules that I am currently using are good for prototyping, but not for a real world scenario I feel. I am very well aware of the beaglebone, rpi, etc. They are good for playing around, not for actual deployment. I want to move past that, all I was suggesting is that I am already working on those things. Thats all."
A $1000 board for thermal imaging sounds weird for 'actual deployment' which I'm guessing you mean as deployment in the field. What was the processing limitation you're going to hit with lower-cost plain Cortex-A and a smaller FPGA if needed, in actual deployment that would need a $1000 board?
"Also, as I mentioned, I have worked on a multitude of projects, not just that one. There are other suitable projects that I am working on as well and I feel I will be able to write a good and extensive review."
Great.. any hints on these projects?
"Lastly, yes I am aware of the pricing of this board. It won't be a part of the deployable solution, I intend to use it more for R and D and applications where I need a more powerful hardware. "
What R&D applications?
I'm only questioning this, because about five years ago an FPGA board got reviewed incompetently, and the guy scrapped the PCB - he tore it down!. Not saying that will be repeated by you, but when you raised what you were planning to do, it rang alarm bells and made me wonder what projects you have done with Xilinx or FPGAs. Given that there may only be one or a few RoadTests for this board, it would be nice to see a decent RoadTest is all I'm interested in seeing.
I understand. But it seems to me that you don't. I will try to sound clear.
I am doing some projects in thermal imaging, that have been restricted to cheaper sensors and boards because in India we don't really have much of funds to actually try and create something innovative or new. I wanted to develop a system, similar to FLIR's MSX technology. That was not possible with the infra I had so I had to resort to limited applications. With this board, it will be like the sky is the limit.
In India, we do not receive much of funding to develop some novel ideas or products easily. It is a real grind and everyone needs returns, and that to soon. I wanted to develop systems that use heavy Image processing in situations where multiple cameras wirelessly send their feeds over a frequency band. Like railway stations or agricultural farms.
A 1000$ board might sound weird for actual deployment yes, but not if it works for 10 people together, and they share it's cost.
As far as my background in fpgas and xilinx is concerned, it is only academical and basic, not even close to what i have done in embedded sbcs and controllers, but i am a fast learner and wanted to try these things that i mentioned in earlier paras on an fpga, since microcontrollers are not always sufficient.
Anyways, it's rscasny decision, all i wanted to mention is that i am seriously interested and if he finds me suitable he can surely give it a chance, i promise not to disappoint if he does. That's all I can say.
3 of 3 people found this helpful
thebluephoenix I'd suggest that it would be wise to finish your existing road test (and impress by doing it well) before taking on something this ambitious. Also, I believe that using a road test board as part of a work project can add some complications and isn't recommended.
After attending yesterday's webinar I'm really impressed by the board but more sure that I wouldn't be able to provide what Avnet would be after in a road test. I'd suggest that anyone hoping to road test this take a look at the webinar recording first Embedded Video Processing Using the UltraZed-EV
I was present during the webinar, it was 3 in the night in India, but i was really interested in knowing more about the board.
I'll do my road test well, and as far as using the ultrazed for one of my project is concerned, it isn't any corporate or backed project. It is a personal project, that i i just wish to deploy in a practical environment. The companies i talked about just sponsor my hardware similar to this case, to be used in one of my applications. I try them, send my reviews and thats it.
As I said, I just mentioned I am interested in this roadtest, i won't disappoint if someone puts faith in me. The rest is for rscasny to decide, i have no problems if someone more experienced gets selected for this roadtest, i'll be looking forward to it.
I too was talking about that roadtest. Knowing my speed, and the time I have to spare, I meant to say that I can do one more roadtest with it. I am already 30% done with it.
rscasny, a beasty board with hardcore features for high-end embedded video processing requirements and I would really love to see this in action. Thinking of an Advanced Driver Assistance System with video analysis based situational awareness application. However, I am still a rookie to Zedboard platform and am currently learning it, and the implementation would thus take additional time. So, considering my inadequate experience, I have to back off.
Due to time constraints I really can't apply for this beast of a board. Will drop in a question here though -
In the Video Processing module I am doing in college, we have been taught there are two big stages in the video pipeline - Compression(reducing the size of a video file) and Post-processing (things like Bayesian probability and motion estimation [Stochastic math]). I am sure the VCU can do (de/)compression, but what about the post-processing bit? Where does an FPGA come in here? I have done quite a lot of work on FPGAs so hit me with all the jargon you guys got!
The VCU is a hardened HECV. H265 decode encoder so if you're familiar with Video codecs you'll find that motion estimation is an inherent part of the algorithm. On H265 search for P-Frames , i-Frames to get an idea.
I am considering to roadtest the UltraZed-EV board. The UltraZed-EV is a highly flexible, rugged, System-On-Module based on the Xilinx Zynq UltraScale+ MPSoC. It's ideal for embedded video processing system development.
In order to roadtest this board, I've discussed with the sponsor what kind of roadtester he needs and any specific requirements as well. Out of those discussions, I've come to the conclusion I need a special roadtester -- someone who has video experience and has the ability to write a quality technical report.
Specifically, these are the things the sponsor is looking for in a roadtester:
- Video Centric (preference to designs using interfaces without an Avnet reference design)
- External pieces (Video/PCIe/SATA/SFP+/etc.) should be COTS and readily available
- Utilize VCU (optional)
- Prefer staying within bounds of VCU TRD (see point 1)
- Have fun, show something exciting and new! (AI/ML?)
If you are interested and have these requirements, please leave a comment and I'll get back to you. Or, you can message me at rscasny
(BTW: If you are new to roadtesting, and are selected, you would receive the UltraZed-EV at no charge to you. I'll give you more info later.)
RoadTest Program Manager