Metadata of recognised object at any given time or duration

[pitoszud]

Accurately counting recognized objects and producing a metadata within a defined timeframe or at any given moment would significantly enhance many applications.

For example, in an Android Compose application for counting passing cars using object recognition, simply drawing a rectangle on the screen and checking for overlaps within that region presents several limitations:

Simultaneous entries: Multiple cars entering the region concurrently would lead to inaccurate counts.
Static objects: Vehicles stopped in traffic would be continuously counted.
Object movement: A car reversing within the region would be counted multiple times.
To address these issues, a more robust approach would involve the model tracking the first occurrence of each object and storing relevant metadata. This metadata could include:

Timestamp of entry/exit: Precisely record the time an object enters and leaves the defined area.
Duration of recognition: Track the amount of time an object is within the region.
Object dimensions: Record the size and shape of the recognized object.
This enhanced tracking capability would significantly improve the accuracy and reliability of object counting applications.

Existing systems like YOLOv8 and YOLO11 already incorporate some of these features, providing a foundation for building more sophisticated object tracking and counting solutions.