Create object detection library

[duwudi]

There is a range of off-the-shelf ONNX object detection models available here. From research, it seemed obvious to go with a YOLO-based model as they have the best performance. Descriptions:

YOLOv3:

A deep CNN model for real-time object detection that detects 80 different classes. A little bigger than YOLOv2 but still very fast. As accurate as SSD but 3 times faster.

Tiny YOLOv3:

A smaller version of YOLOv3 model.

YOLOv4:

Optimizes the speed and accuracy of object detection. Two times faster than EfficientDet. It improves YOLOv3's AP and FPS by 10% and 12%, respectively, with mAP50 of 52.32 on the COCO 2017 dataset and FPS of 41.7 on a Tesla V100.

I have tested both Tiny V3 and V4 on a pi-top [4] with 640 x 480 resolution video feed, results:

Tiny YOLOv3: 1.2 FPS
YOLOv4: 0.11 FPS

Unfortunately, there isn't an ONNX model for Tiny YOLOv4. We could create it and it would be faster and more accurate than Tiny v3 according to this