Analysis
A configuration can contain any number of analysis tasks. For example, you can run face detection and object recognition at the same time.
The following example includes a single analysis task named OCR
. The task uses the OCR
analysis engine:
[Session] Engine0=Ingest Engine1=OCR [OCR] Type=ocr Input=Default_Image
An analysis engine accepts input of a particular type. For example, the OCR engine requires an image track, and the SpeakerID engine requires an audio track. The analysis engine only processes records from the track specified by the Input
configuration parameter. In this example, the OCR engine processes the Default_Image
track produced by the ingest engine. The default value of the Input
parameter is Default_Image
for engines that require images and Default_Audio
for engines that require audio, so in many cases you do not need to include the Input
parameter in the configuration.
Some analysis engines require more complex input. For example, the face recognition analysis engine requires records that contain an image but also region information that specifies the position of the face in the image. The region information is not available in the Default_Image
track from the ingest engine so you cannot use that track as the input and you must set the Input
configuration parameter. The region information is provided by a face detection analysis task. The DataWithSource
track produced by face detection includes the location of each face in every frame and the corresponding source images.
In the following example the input of the face recognition task is the DataWithSource
track produced by the face detection task:
[Session] Engine0=Ingest Engine1=FaceDetect Engine2=FaceRecognize [FaceDetect] Type=FaceDetect ... [FaceRecognize] Type=FaceRecognize Input=FaceDetect.DataWithSource ...
The response to action=ListEngines
describes the type of input required by each analysis engine. It also describes the output tracks that are produced. To be used as the input for a task, a track must provide at least the required record types. For more information about the output tracks produced by Media Server engines, refer to the Media Server Reference.