search

bitsy-ai / rpi-object-tracking

Object tracking tutorial using TensorFlow / TensorFlow Lite, Raspberry Pi, Pi Camera, and a Pimoroni Pan-Tilt Hat.
https://medium.com/@grepLeigh/real-time-object-tracking-with-tensorflow-raspberry-pi-and-pan-tilt-hat-2aeaef47e134
MIT License
184 stars 68 forks source link

Use a USB Webcam? #39

Closed kindofausername closed 4 years ago

kindofausername commented 4 years ago

Description

Is it possible to use a regular USB Webcam? What changes do I have to make to make it run?

Additional question: When I run a coral model on object detection I get a max 8 FPS. Would that be faster using a Pi Cam?

Regards, Armin

leigh-johnson commented 4 years ago

Hey @arminf82! I haven't tried a USB web cam. The performance is definitely going to depend on the camera's frame rate, USB 2 vs 3, and how efficiently the camera's driver can serialize data.

The PiCamera uses a MIPI serial interface, which is specifically designed for high-speed image data transfer. My gut feeling is that it'll have a higher bitrate on the Pi than most USB cameras. If I were trying to compare the two, I'd write a tiny program that benchmarks how fast each method can read/write to a ring buffer. https://www.mipi.org/specifications/csi-2

You could also use cprofile to see if the inference loop is waiting on buffer being available to read.

Here's the line to replace if you implement a USB image stream. Replace this .read() method with one that takes the most recent frame from your buffer, then zeroes the buffer. https://github.com/leigh-johnson/rpi-deep-pantilt/blob/master/rpi_deep_pantilt/detect/camera.py#L93

leigh-johnson commented 4 years ago

Closing for now, so it's easier for me to see issues in need of triage at a glance. Good luck with this!

kindofausername commented 4 years ago

I bought a PiCam v2. I had two different USB Cams and both of them were much slower. Thank you.