Notifications
Clear all
Topic starter
2022-01-04 4:45 pm
I'm primarily a software developer, working in the Eindhoven region of the Netherlands. I have also dabbled in some electronic design. My main platforms are Linux desktops, raspberrypi's, Jetson series boards Nano, Xavier NX and Xavier AGX, lorawan client devices. Actually lots of things. I program in Java, Python, shell script, perl JavaScript, react, perl, lots of languages really.
I keep myself busy with my open source project "StalkedByTheState" a yolov3 and yolov4 powered home and business security systen and everything that hangs around this. I just started a YouTube channel ( https://www.youtube.com/channel/UCXn7Z37_xwuxLPpcPTtdNRQ) around a month ago. It's very new and I'm sure that shows at the moment. The github repos are at https://github.com/hcfman
Currently, I'm trying to introduce people to my new open source project, which is at the brand new stage, meaning very few people have used it. Right now only people I know personally I think. But I'll open a topic to introduce this to people and explain what it can do. I believe it crosses the barrier from being interesting to being very practical to use, now is my job to convince people of that 🙂 In fact, it's been in use by people with the CNN computer vision enhancements for around two years though, but not released till recently.
Anyway, nice to meet you all,
Cheers,
Kim Hendrikse
DroneBot Workshop reacted
2022-01-05 6:44 am
Hi Kim, Always nice to meet a fellow Dutchy on the board. 🙂
I'll have a look at your project later today, sounds very interesting!
Topic starter
2022-01-05 8:56 am
@markasread Dankjewel!
In my small village where I live in Limburg, 4x homes are using this software for their home security. A fifth one is on the way when we can get hold on a Jetson Xavier NX and he is a former police man of 30 years. I take that as a good sign that people like to continue using it once they start. I've been using and evolving the non-AI fronted version for more than 10 years as well as some other people. (Meaning using other forms of sensors, such as PIR sensors or hardware hacked PIR sensors to turn them into Lorawan enabled PIR sensors [I'll do a tutorial on this sometime]).
But the latest changes are the most interesting. I don't access the yolov3 or yolov4 models directly. I've modified the code to expose them as a websocket service, you throw an image to the service and you get a classification back. This makes it more flexible as you can use more than one AI model at the same time.
In the context of a robot this also means that you could in principle throw binaries images to a remote version of the software running with a high powered GPU and get classifications back in real time. Fast enough to use as part of robot navigation for example.
On an rtx 2080ti it can process 30x frames per second. I'm guessing on a an rtx 3080ti this would be 60 frames per second. In any case, it's trival to access to remote network based I recognition in python, the example tools included in the package show how this is done. The only complexity is that you have to use asynchronous programming. But apart from that I think it could open perspectives for controlling your robot. If a remote link for computer vision is appropriate for the use case. Certainly the bandwidth is really low so there's no technical challenge in using this, so long as you can maintain a stable network connection.
Kim
Inst-Tech reacted