Object Detection Using CodeProject.AI

We can use CodeProject.AI to detect objects in images. I'm using it with my Blue Iris security system so that I only see notifications when an object I'm interested in appears in a video feed - such as a person. We look at some configuration options to get the best performance.

[@user-vnatury]

The first blogger, that i watch on 0.75 instead of 1.25)

[@thierrybo6304]

😂

[@codegearguru]

Yes, I talk pretty fast. I probably need to relax a bit and slow down.

[@sanderfrigolagroen]

It's a big compliment, the 0.75 speed comment. :) @@codegearguru

[@daslolo]

"Glances back over his shoulder suspiciously"

[@dualnamehype]

Hi I've just noticed you were using a amd gpu i was under the impression that was not possible how have u done that?

[@codegearguru]

This is using the YOLOv5 .NET, which will work with AMD GPUs. I've since changed to a NVidia GPU (GTX 970) for H264 support for Blue Iris. But the .NET implementation is faster and you should probably use it if it works.

[@dualnamehype]

@@codegearguru thanks for the response il look into that

[@S.C.A.M.B.E.R.]

Hi, do you think 2MP (turret and bullet) dahua or hikvision camera would be enough for AI recognition for faces, pets, and object, not including ALPR.

[@codegearguru]

That's going to be a "depends". Potentially yes. I've not done much face recognition and I think you need a fairly close image to make this work. You can probably get ALPR working as well if you wanted. It's probably a matter of trying it out. I haven't used Dahua cameras, but I have a couple of HikVision cameras that are many years old and still going strong. Wouldn't recommend Reolink for Blue Iris, but Amcrest seem to work fine.

[@michaelpoczynek]

Thanks for making this very well-done and informative video. I have a question: Facial Recognition is not working. Do you have a checklist of things to be in place? Blue Iris & Code Project are both working and talking. Face Processing is ON. It seems like BIue Iris is not sending AI any faces.

[@codegearguru]

I haven't done much with face detection as yet, all of my cameras are wide shots, no real way of getting close-ups on faces.

[@michaelpoczynek]

@@codegearguruThank you.

[@yossishi221]

hi i got lost on benchmark what is better?? more operational in seconds? or less ? thanks yossi

[@codegearguru]

You want as many operations/detections per second as you can, this will allow you to do far better object detection as it can do it in more frames of a video.

[@yossishi221]

@@codegearguru thanks the reason i ask is because on the last update on ai i have problem with gpu i have asus gtx 1060 6gb turbo and when it on yolov v5 3.1 its use only cpu also cpu on yolo v5 6.2 only gpu is on yolo v5 net so how can it be that it not use local gpu ??? thanks Yossi

[@Joe99]

Have you experienced a decrease in false detections while using this?

[@codegearguru]

Absolutely, it filters out moving trees and cloud shadows - assuming there isn't a car in the frame as well.

[@Will-xk4nm]

@@codegearguru I had to disable car detection for my driveway cam precisely because any motion would cause it to trigger and subsequently AI would spot my car and send the alert. I'd really love to get individual vehicle detection in the same way we handle facial detection in CPAI so my own vehicle doesn't send 50 alerts a day.

[@techniciendelextreme8855]

Hello, thank you for sharing your experience. I use a GTX 1050, and I have made the same observation! YOLO net is better than CUDA! I would have liked to hear about your experience with face detection. With an i5 9500 CPU, I'm running at 150ms, but with CUDA, it takes 700ms...

[@codegearguru]

I've only briefly played with face detection. Most of my cameras don't get close up views of faces so it's a bit challenging. I should set up a couple of other cameras for this purpose.