I executed the darknet Yolov3-tiny posted YOLO: Real-Time Object Detection with a Raspberry Pi camera and got around 13 fps after the IA. However when I execute the same with the e-CAM50 CUNX camera from econsystems with the same image resolution and same algorithm I get only about 1.7 fps.
Does anybody know what the problem could be? Is it related to the driver from econsystems? Is there a way to solve it and get a higher speed?
Have you check the camera cap by the v4l2-ctl to confirm it. Also confirm if set the sensor output size are the same.
Yes, when I check the v4l2-ctl the fps of the camera varies depending on what it is capturing. I mean with the cover on it only got 7 fps and with the cover off it got from 24 to 65 fps (this one is the maximun the camera can give for the resolution I was trying). This behavior is weird but I guess that it is an internal configuration that I am not interested in and I hope can be modified.
However my biggest problem comes when I run the IA to the cameras input, because doing the same as I did with the Raspberry pi camera the fps in lower.
I don’t think that’s a problem. The frame rate drop because the AE alg need adjust to get longer time of exposure to get better preview.
You can try the v4l2-ctl to check if any frame rate control to set it as fixed frame rate to try, otherwise you may need consult with vendor.