META: Object detection

[magsol]

Ideally, we'd like to perform real-time object detection. Whether this is possible on a Pi 3B+ has yet to be determined, but we'll proceed for now as though it is.

In particular, we want to detect:

• people (full body)
• front faces
• cars
• birds, bugs, deer, cats
• other animals

OpenCV has some built-in Haar Cascades pre-trained weights, in particular:

• haarcascade_frontalface_default.xml
• haarcascade_fullbody.xml

(these are located at ~/.local/lib/python3.9/site-packages/cv2/data)

But beyond that, we'll need a custom object detector.

[magsol]

First, some downloads.

yolov4-tiny weights: https://github.com/AlexeyAB/darknet/releases/download/yolov4/yolov4-tiny.weights

yolov4-tiny config: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov4-tiny.cfg

Classes can be found packaged with OpenCV in the examples: /path/to/opencv/samples/data/dnn/object_detection_classes_yolov3.txt