Rekognition Streaming Video events and Custom Labels

Question

Hi,

i am trying to configure  Rekognition with Streaming Video Events but i am having a hard time when trying to use custom labels. Works fine with Face detection but i want to detect specific "labels" in a live video.

I have created a KVS, configured the SDK and can stream video from my laptop and IP camera.

But, if try to create a Stream processor with custom labels i keep getting an error
`An error occurred (InvalidParameterException) when calling the CreateStreamProcessor operation: Invalid label(s) specified for ConnectedHome: "XXXXX", "YYYYY"`

Is there a limitation in use cases so that Streaming Video Events can only be used with Person, Pet, All labels?

I know that a workaround is to use a kinesis stream, lambda, fetch frames from the video , store them in S3 and Rekognition will do the rest.

But i was hoping that could do the same with live video and Streaming Video Events.

Thanks

AWS-User-1104020 · Answer

Dear Nikitas,

Thanks for reaching out.

CreateStreamProcessor API can be used to create StreamProcessors with two different settings in Amazon Rekognition: detecting faces (FaceSearch option in Settings) and detecting labels (ConnectedHome option in Settings).

Rekognition Streaming Video Events currently only offers label detection on one of the following labels: PERSON, PET, PACKAGE, ALL for ConnectedHome option in Settings.

You can refer to the below documentation for more details:
* CreateStreamProcessor: https://docs.aws.amazon.com/rekognition/latest/APIReference/API_CreateStreamProcessor.html
* Label detection operations for streaming video events: https://docs.aws.amazon.com/rekognition/latest/dg/streaming-labels-detection.html
* Detecting labels in a video: https://docs.aws.amazon.com/rekognition/latest/dg/labels-detecting-labels-video.html
* Detecting custom labels: https://docs.aws.amazon.com/rekognition/latest/dg/labels-detecting-custom-labels.html

Regards,

AWS Team

Rekognition Streaming Video events and Custom Labels

Conteúdo relevante