Hi, I have a pipeline that first runs an object detection neural network and then feeds its detections to an object tracker node. Since the network doesnt achieve \~30 FPS I would like to get more frequent updates from the tracker based on short term predictions. In the [documentation](https://docs.luxonis.com/projects/api/en/latest/components/nodes/object_tracker/) I've found the statement `… tracker can provide tracklets when there as no inference, so the wholse system can run at 30FPS`*.* As suggested in the documentation, I've tried using the `dai.TrackerType.SHORT_TERM_KFC` to get more frequent updates, but to no avail. Are there any additional settings that need to be set here?

Hi Stn73! To get more frequent updates from the tracker, you should indeed use the `dai.TrackerType.SHORT_TERM_KCF` or `dai.TrackerType.SHORT_TERM_IMAGELESS` as these allow tracking objects on frames where object detection was skipped, by extrapolating object trajectory from previous detections. However, there are a few other settings that you might need to check: 1. Make sure you are using the Object Tracker in combination with a detection network and an image frame source, such as a mono/color camera or xlinkIn node. 2. Check the maximum number of objects you are tracking. The Object Tracker node can track up to 60 objects at once for `SHORT_TERM_KCF` and up to 1000 for other tracking methods. If there are more than 60 objects to track, the firmware might crash. 3. Ensure that the detection labels you are tracking are correct. For example, if you are tracking only persons, you should set the detection labels to track as `[15]`. 4. You might also want to consider reducing the frame rate to match, or be just below, the frame rate of the neural inference if your model cannot be executed at the full 30FPS. This can help to regain synchronization. Here is an example of how to set up the Object Tracker: ```py pipeline = dai.Pipeline() objectTracker = pipeline.create(dai.node.ObjectTracker) objectTracker.setDetectionLabelsToTrack([15]) # Track only person objectTracker.setTrackerType(dai.TrackerType.SHORT_TERM_KCF) objectTracker.setTrackerIdAssignmentPolicy(dai.TrackerIdAssignmentPolicy.SMALLEST_ID) ``` Sources: - [Object Tracker Node](https://docs.luxonis.com/projects/api/en/latest/components/nodes/object_tracker/) - [Object Tracker Usage](https://docs.luxonis.com/projects/api/en/latest/components/nodes/object_tracker/#usage) - [Supported Object Tracker Types](https://docs.luxonis.com/projects/api/en/latest/components/nodes/object_tracker/#supported-object-tracker-types) - [Maximum Number of Tracked Objects](https://docs.luxonis.com/projects/api/en/latest/components/nodes/object_tracker/#maximum-number-of-tracked-objects) - [Reducing the Camera Frame Rate](https://docs.luxonis.com/pages/faq/#reducing-the-camera-frame-rate) - [Object Tracker Properties](https://docs.luxonis.com/projects/api/en/latest/references/python/#depthai.ObjectTrackerProperties.maxObjectsToTrack) Hope this helps! :) Best, Luxonis BOT

Hi @"Stn73"#p14161 - Try using SHORT_TERM_IMAGELESS - try linking the camera directly to the trackerFrame and detectionFrame See if it improves your FPS. Also make sure the camera is able to produce frames fast enough. Thanks, Jaka

Getting Updates from Object Tracker in between Detections

Stn73

Hi, I have a pipeline that first runs an object detection neural network and then feeds its detections to an object tracker node. Since the network doesnt achieve ~30 FPS I would like to get more frequent updates from the tracker based on short term predictions.

In the documentation I've found the statement … tracker can provide tracklets when there as no inference, so the wholse system can run at 30FPS. As suggested in the documentation, I've tried using the dai.TrackerType.SHORT_TERM_KFC to get more frequent updates, but to no avail. Are there any additional settings that need to be set here?

jakaskerl

Hi Stn73

Try using SHORT_TERM_IMAGELESS
try linking the camera directly to the trackerFrame and detectionFrame

See if it improves your FPS.
Also make sure the camera is able to produce frames fast enough.

Thanks,
Jaka

Stn73

Thanks, I will try that out!