I understand that you want to manage the use of detections with a tracker, especially when dealing with varying inputs over time. Initially, providing a nonempty cell array of detections to the tracker is essential for its setup and input validation. This step is crucial for the tracker to properly initialize and prepare for subsequent tracking operations.
Once the tracker has been initialized and is in a "locked" state, it's possible to pass an empty cell array of detections in subsequent calls. This approach is particularly useful when detections are sparse or when there are frames without any detectable objects. The pattern for handling this can be seen in below examples link-
