Track Detections Node

T

AI/ML/Tracking

Follows detected objects across the frames of one camera and gives each a stable track id (ByteTrack, with BoT-SORT appearance matching when detections carry embeddings). Run it once per frame: the tracker is kept in memory per user, board, node, Camera pin and Session pin between runs and uses frame timestamps for motion and expiry. Track ids are only unique per tracker instance, named by each track's tracker_id: a new Camera or Session pin value, 10 minutes without frames, eviction when too many trackers are open, or a process restart starts a new tracker with a new tracker_id (a session carried by the detections does not). Deployments that spread runs over several processes keep one independent tracker per process.

tracking_track_detectionsonnx

Inputs13

Outputs3

Security exposure8/10

Packageonnx

Ratings

Scores range from 0 to 10. Higher values mean more impact, exposure, or operational weight.

SecurityAttack surface and exposure impact.

8/10High

PrivacyPotential sensitivity of processed data.

8/10High

PerformanceRuntime or resource pressure.

9/10High

GovernancePolicy, audit, or compliance impact.

7/10High

ReliabilityOperational stability considerations.

7/10High

CostExternal or compute cost impact.

10/10High

Input Pins

13

Input

Execution

exec_in

Initiate Execution

Detections

Struct Array

detections

Boxes detected in this frame. Embeddings from Extract Appearance are used for appearance matching when present. Elements in the lost state (predicted boxes of lost tracks fed back in) are ignored.

SchemaBoundingBoxBoundingBoxBoundingBox7 fields

x1number:floatrequired

format float

y1number:floatrequired

format float

x2number:floatrequired

format float

y2number:floatrequired

format float

scorenumber:floatrequired

format float

class_idxinteger:int32required

format int32

class_namestring | null

Default []

Camera

String

camera_id

Camera the frame comes from. Empty keeps the camera carried by the detections, or the last one seen. Every value of this pin gets its own tracker, so set it when one node tracks several cameras.

Session

String

session_id

Stream session of the camera. Empty keeps the session carried by the detections, or the last one seen. A new value of this pin starts a fresh tracker with a new tracker_id.

Timestamp (ms)

Integer

timestamp_ms

Frame capture time in Unix milliseconds; 0 uses the current time, never earlier than the last processed frame. A frame up to Max Lost older than the last processed one is skipped; one further back is taken as a clock reset that drops all tracks.

Default 0

High Threshold

Float

high_threshold

Detections scoring at least this take part in the first association

Default 0.5

Range 0 to 1

Low Threshold

Float

low_threshold

Detections scoring at least this, but below the high threshold, can only keep existing tracks alive

Default 0.1

Range 0 to 1

New Track Threshold

Float

new_track_threshold

Minimum score for an unmatched detection to start a new track

Default 0.6

Range 0 to 1

Match Threshold

Float

match_threshold

Maximum matching cost (1 - IoU × score, or appearance distance) of the first association; higher matches more loosely

Default 0.8

Range 0 to 1

Max Lost (ms)

Integer

max_lost_ms

How long a track that lost its object can be re-acquired with the same id

Default 2000

Min Appearance Similarity

Float

min_appearance_similarity

Minimum cosine similarity of embeddings for appearance to count as a match

Default 0.5

Range 0 to 1

Class Aware

Boolean

class_aware

Never match a track with a detection of another class

Default true

Include Lost

Boolean

include_lost

Also output tracks that lost their object. Their boxes are Kalman predictions, not detections: Extract Appearance and Associate Entities treat them as passive, and this node ignores them when they are fed back in.

Default false

Output Pins

3

Output

Execution

exec_out

Done with the Execution

Tracks

Struct Array

tracks

Confirmed tracks matched in this frame (plus lost tracks if enabled), sorted by track id. Track ids are unique per tracker_id.

SchemaTrackedObjectTrackedObjectTrackedObject20 fields

x1number:floatrequired

format float

y1number:floatrequired

format float

x2number:floatrequired

format float

y2number:floatrequired

format float

scorenumber:floatrequired

format float

class_idxinteger:int32required

format int32

class_namestring | null

embeddingArray<number:float>

Smoothed appearance embedding of the track; empty when no embeddings were supplied

default []

itemsnumber:floatarray item

format float

camera_idstring

default ""

session_idstring

default ""

tracker_idstring

Tracker instance that issued `track_id`. A new tracker (new Camera/Session pin value, 10 minutes without frames, eviction or restart) gets a new one; a clock reset keeps it

default ""

track_idinteger | null

Track id, unique within one tracker instance

format uint64default nullmin 0

timestamp_msinteger:int64

Timestamp of the frame this track state belongs to, in Unix milliseconds

format int64default 0

detection_indexinteger | null

Index of the matched detection in this frame's input, or null for lost tracks

format uint32default nullmin 0

stateTrackState

default "tracked"

variant 1constvariant

Matched to a detection in the latest frame

const "tracked"

variant 2constvariant

Not matched recently; kept alive for re-identification until it expires

const "lost"

hitsinteger:uint32

Number of frames the track was matched in

format uint32default 0min 0

first_seen_msinteger:int64

format int64default 0

last_seen_msinteger:int64

format int64default 0

+2 more fields

Default []

Skipped

Boolean

skipped

True when the frame was older than the last processed one and was ignored

Node Info

Internal name

tracking_track_detections

Category

AI/ML/Tracking

Version

1