Open topic with navigation
Analysis Task Output Tracks
The events that occur in video usually span multiple frames. For example, a person, object, or logo might appear on screen and remain there for seconds or minutes. Media Server analysis tasks run analysis on single frames, but also follow events across multiple frames.
Analysis tasks can produce the following tracks:
- Data tracks contain a record for every frame of an event. For example, the data track from a face detection task contains one or more records for every frame in which a face is detected (Media Server creates multiple records when a frame contains more than one face). If a person remains in the scene for several seconds, this track could contain hundreds of records that represent the same face.
- DataWithSource tracks are similar to data tracks because they contain one or more records for every frame of an event. However, each record also includes the image analyzed by the task to produce the record.
- Result tracks contain a single record for each event. The result track summarizes the results of analysis for each event. For example, the result track from a face detection task contains a single record for each detected face. Each record has a start time, a peak time, a duration, and an end time, and can span many frames. Result tracks are, by default, used as input tracks for output tasks that you configure.
- ResultWithSource tracks are similar to result tracks because they contain a single record for each event. However, each record also includes the source image that produced the best result. For example, when you run face recognition the frame with the highest confidence score is added to the record. This frame corresponds to the peak time in the record timestamp.
- Event tracks contain records that describe the beginning or end of events in the video. For example, the event track for a face detection task contains a record when a face appears in the video, and another record when the face disappears from the scene. Event tracks are, by default, used as input tracks for output tasks that you configure.
The following diagram shows how Media Server creates records (represented by blue squares) when a person appears in a video.
- The Face Detection task creates records in the
Event track when the person appears and disappears from the scene.
- The task creates one record in the
DataWithSource tracks for each analyzed frame. If there are multiple people in the scene, Media Server creates multiple records in the
DataWithSource tracks for each frame.
- The task creates a single record in the
ResultWithSource tracks for each event (in this case a detected face). This record spans the event and summarizes the analysis results.
For information about the tracks that are produced by Media Server tasks, and the data contained in each track, refer to the Media Server Reference.