Object Detection
Real-time object detection built with TensorFlow. Identifies and labels objects from a live camera feed.
the problem
Most object-detection tutorials stop at running a model against a static image. Getting it to run against a live camera feed without lagging, and keeping the label set correct and scoped, is a different problem.
approach
Loads a pretrained SSD MobileNet V2 from TensorFlow Hub once, then runs an OpenCV capture loop: read a frame, convert BGR to RGB, run inference, draw the results, repeat. The label map is deliberately trimmed to a small, correct subset rather than the full 90-class COCO set, since a smaller label set is easier to verify by hand for a class assignment.
architecture
TensorFlow Hub model
SSD MobileNet V2, loaded once at startup and reused across every frame instead of reloading per inference.
OpenCV capture loop
Reads live frames from a camera device, handles the BGR-to-RGB conversion TensorFlow expects, and runs inference in the same loop.
Trimmed label map
A small, hand-picked label set instead of the full COCO taxonomy -- kept deliberately narrow and verifiably correct.