Amazon Kinesis Video Streams Parser Library

License

This library is licensed under the Apache 2.0 License.

Introduction

The Amazon Kinesis Video Streams Parser Library for Java enables Java developers to parse the streams returned by GetMedia calls to Amazon Kinesis Video. It contains:

• A streaming Mkv Parser called StreamingMkvReader that provides an iterative interface to read the MkvElements in a stream.
• Applications such as OutputSegmentMerger and FragmentMetadataVisitor built using the StreamingMkvReader .
• A callback based parser called EBMLParser that minimizes data buffering and copying. StreamingMkvReader is built on top of EBMLParser
• Unit tests for the applications and parsers that demonstrate how the applications work.

Building from Source

After you've downloaded the code from GitHub, you can build it using Maven. Use this command: mvn clean install

Details

StreamingMkvReader

StreamingMkvReader which provides an iterative interface to read MkvElements from a stream. A caller calls nextIfAvailable to get the next MkvElement. An MkvElement wrapped in an Optional is returned if a complete element is available. It buffers an individual MkvElement until it can return a complete MkvElement.
The mightHaveNext method returns true if there is a chance that additional MkvElements can be returned. It returns false when the end of the input stream has been reached.

MkvElement

There are three types of MkvElement vended by a MkvStreamReader:

• MkvDataElement: This encapsulates Mkv Elements that are not master elements and contain data.
• MkvStartMasterElement : This represents the start of a master Mkv element that contains child elements. Child elements can be other master elements or data elements.
• MkvEndMasterElement : This represents the end of master element that contains child elements.

MkvElementVisitor

The MkvElementVisitor is a visitor pattern that helps process the events in the MkvElement hierarchy. It has a visit method for each type of MkvElement.

Visitors

A GetMedia call to Kinesis Video vends a stream of fragments where each fragment is encapsulated in a Mkv stream containing EBML and Segment elements. OutputSegmentMerger can be used to merge consecutive fragments that share the same EBML and Track data into a single Mkv stream with a shared EBML and Segment. This is useful for passing the output of GetMedia to any downstream processor that expects a single Mkv stream with one Segment. Its use can be seen in OutputSegmentMergerTest

FragmentMetadataVisitor is a MkvElementVisitor that collects the Kinesis Video specific meta-data (such as FragmentNumber and Server Side Timestamp ) for the current fragment being processed. The getCurrentFragmentMetadata method can be used to get the current fragment's metadata. Similarly getPreviousFragmentMetadata can be used get the previous fragment's metadata. The getMkvTrackMetadata method can be used to get the details of a particular track.

ElementSizeAndOffsetVisitor is a visitor that writes out the metadata of the Mkv elements in a stream. For each element the name, offset, header size and data size is written out. The output uses indentation to indicate the hierarchy of master elements and their child elements. ElementSizeAndOffsetVisitor is useful for looking into Mkv streams, where mkvinfo fails.

CountVisitor is a visitor that can be used to count the number of Mkv elements of different types in a Mkv stream.

CompositeMkvElementVisitor is a visitor that is made up of a number of constituent visitors. It calls accept on the visited MkvElement for each constituent visitor in the order in which the visitors are specified.

FrameVisitor is a visitor used to process the frames in the output of a GetMedia call. It invokes an implementation of the FrameVisitor.FrameProcessor and provides it with a Frame object and the metadata of the track to which the Frame belongs.

CopyVisitor is a visitor used to copy the raw bytes of the Mkv elements in a stream to an output stream.

ResponseStreamConsumers

The GetMediaResponseStreamConsumer is an abstract class used to consume the output of a GetMedia* call to Kinesis Video in a streaming fashion. It supports a single abstract method called process that is invoked to process the streaming payload of a GetMedia response. The first parameter for process method is the payload inputStream in a GetMediaResult returned by a call to GetMedia. Implementations of the process method of this interface should block until all the data in the inputStream has been processed or the process method decides to stop for some other reason. The second argument is a FragmentMetadataCallback which is invoked at the end of every processed fragment. The GetMediaResponseStreamConsumer provides a utility method processWithFragmentEndCallbacks that can be used by child classes to implement the end of fragment callbacks. The process method can be implemented using a combination of the visitors described earlier.

MergedOutputPiper

The MergedOutputPiper extends GetMediaResponseStreamConsumer to merge consecutive mkv streams in the output of GetMedia and pipes the merged stream to the stdin of a child process. It is meant to be used to pipe the output of a GetMedia* call to a processing application that can not deal with having multiple consecutive mkv streams. Gstreamer is one such application that requires a merged stream.

Example

KinesisVideoExample

KinesisVideoExample is an example that shows how the StreamingMkvReader and the different visitors can be integrated with the AWS SDK for the Kinesis Video. This example provides examples for

• Create a stream, deleting and recreating if the stream of the same name already exists.
• Call PutMedia to stream video fragments into the stream.
• Simultaneously call GetMedia to stream video fragments out of the stream.
• It uses the StreamingMkvParser to parse the returned the stream and apply the OutputSegmentMerger, FragmentMetadataVisitor visitors along with a local one as part of the same CompositeMkvElementVisitor visitor.

KinesisVideoRendererExample

KinesisVideoRendererExample shows parsing and rendering of KVS video stream fragments using JCodec(http://jcodec.org/) that were ingested using Producer SDK GStreamer sample application.

• To run the example: Run the Unit test testExample in KinesisVideoRendererExampleTest. After starting the unitTest you should be able to view the frames in a JFrame.
• If you want to store it as image files you could do it by adding (in KinesisVideoRendererExample after AWTUtil.toBufferedImage(rgb, renderImage); )

try {
    ImageIO.write(renderImage, "png", new File(String.format("frame-capture-%s.png", UUID.randomUUID())));
 } catch (IOException e) {
    log.warn("Couldn't convert to a PNG", e);
}

• It has been tested not only for streams ingested by PutMediaWorker but also streams sent to Kinesis Video Streams using GStreamer Demo application (https://github.com/awslabs/amazon-kinesis-video-streams-producer-sdk-cpp)

KinesisVideoGStreamerPiperExample

KinesisVideoGStreamerPiperExample is an example for continuously piping the output of GetMedia calls from a Kinesis Video stream to GStreamer. The test KinesisVideoGStreamerPiperExampleTest provides an example that pipes the output of a KVS GetMedia call to a Gstreamer pipeline. The Gstreamer pipeline is a toy example that demonstrates that Gstreamer can parse the mkv passed into it.

KinesisVideo - Rekognition Examples

Kinesis Video - Rekognition examples demonstrate how to combine Rekognition outputs (JSON) with KinesisVideo Streams output (H264 fragments) and render the frames with overlapping bounding boxes.

KinesisVideoRekognitionIntegrationExample

KinesisVideoRekognitionIntegrationExample decodes H264 frames and renders them with bounding boxes locally using JFrame. To run the sample, run KinesisVideoRekognitionIntegrationExampleTest with appropriate inputs.

KinesisVideoRekognitionLambdaExample

KinesisVideoRekognitionLambdaExample decodes H264 frames, overlaps bounding boxes, encodes to H264 frames again and ingests them into a new Kinesis Video streams using Kinesis Video Producer SDK. To run the sample follow the below steps:

• Goto IAM role, and Create a role for AWS service and select 'Lambda' in 'Choose the service that will use this role'. Click Next
• Click Create Policy' and goto the JSON tab and paste the below. Click 'Review Policy' and specify a policy name to create policy.

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "Editor",
            "Effect": "Allow",
            "Action": [
                "dynamodb:*",
                "logs:CreateLogGroup",
                "logs:CreateLogStream",
                "logs:PutLogEvents",
                "kinesis:Get*",
                "kinesis:List*",
                "kinesis:Describe*",
                "kinesisvideo:*"
            ],
            "Resource": "*"
        }
    ]
}

• Refresh the IAM role page and add the policy created above. Leave rest of the options with default values and enter the role name in last page to create IAM role.
• Create a Lambda from scratch called 'KinesisVideoRekognitionSample' with Java 8 as runtime. For IAM role, choose the role created above.
• Add 'Kinesis' as the trigger for the lambda and select Kinesis Data Streams name in which Rekognition output is emitted.
• Select batch size as 100 and starting position as 'Latest'. 'Add' and 'Save' the changes with "Enable trigger" option checked.
• Run mvn package from 'amazon-kinesis-video-streams-parser-library' folder. Upload './target/amazon-kinesis-video-streams-parser-library-$VERSION-shaded.jar' file to a S3 bucket.
• Click on the lambda name and specify 'Code entry type' as 'Upload file from S3' and specify the above S3 object URL.
• Change the 'Handler' to com.amazonaws.kinesisvideo.parser.examples.lambda.KinesisVideoRekognitionLambdaExample::handleRequest.
• Add below Environment variables. Specify the input KinesisVideo stream name in -DKVSStreamName parameter.
  • key : DISPLAY value : localhost:0.0
  • key : JAVA_TOOL_OPTIONS value : -Djava.awt.headless=true -DKVSStreamName="" -Dlog4j.debug
• Choose the execution role as the IAM role created above. Set 'Memory (MB)' to max value and 'Timeout' as 5 minutes.
• Set Concurrency to '1' as lambda should process fragments in order. Click 'Save'.
• Now start the producer to ingest data into Kinesis Video Streams for which the Rekognition stream processor is configured.
• Lambda will be triggered as soon as the Rekognition stream processor starts emitting records in Kinesis Data Streams. Lambda will also create a new Kinesis Video streams with the input stream name + 'Rekognized' suffix and ingest frames overlapped with bounding boxes which should be viewable in Kinesis Video Streams console.
• To troubleshoot any issues, use Monitoring tab in lambda and click 'View logs in Cloudwatch'.
• NOTE: As this lambda executes resource intense decoding and encoding (using Jcodec which is not optimal https://github.com/jcodec/jcodec#performance--quality-considerations), the new Kinesis Video stream might be delayed significantly.

Release Notes

Release 1.0.9 (Jan 2019)

• Added KinesisVideo Rekgonition Lamba example which combines Rekognition output with KVS fragments to draw bounding boxes for detected faces and ingest into new KVS Stream.
• Added Optional track number parameter in the FrameVisitor to process only frames with that track number.

Release 1.0.8 (Dec 2018)

• Add close method for derived classes to cleanup resources.
• Add exception type which could be used in downstream frame processing logic.
• Make boolean value thread-safe in ContinuousGetMediaWorker.
• Remove extra exception wrapping in CompositeMkvElementVisitor.
• Declare exception throwing for some methods.
• Enabled stack trace in ContinuousGetMediaWorker when there is an exception.

Release 1.0.7 (Sep 2018)

• Add flag in KinesisVideoRendererExample and KinesisVideoExample to use the existing stream (and not doing PutMedia again if it exists already).
• Added support to retrieve the information from FragmentMetadata and display in the image panel during rendering.

Release 1.0.6 (Sep 2018)

• Introduce handling for empty fragment metadata
• Added a new SimpleFrame Visitor to handle video with no tags
• Refactored H264FrameDecoder, so that base class method can be reused by child class

Release 1.0.5 (May 2018)

• Introduce GetMediaResponseStreamConsumer as an abstract class used to consume the output of a GetMedia* call to Kinesis Video in a streaming fashion. Child classes will use visitors to implement different consumers.
• The MergedOutputPiper extends GetMediaResponseStreamConsumer to merge consecutive mkv streams in the output of GetMedia and pipes the merged stream to the stdin of a child process.
• Add the capability and example to pipe the output of GetMedia calls to GStreamer using MergedOutputPiper.

Release 1.0.4 (April 2018)

• Add example for KinesisVideo Streams integration with Rekognition and draw Bounding Boxes for every sampled frame.
• Fix for stream ending before reporting tags visited.
• Same test data file for parsing and rendering example.
• Known Issues: In KinesisVideoRekognitionIntegrationExample, the decode/renderer sample using JCodec may not be able to decode all mkv files.

Release 1.0.3 (Februrary 2018)

• In OutputSegmentMerger, make sure that the lastClusterTimecode is updated for the first fragment. If timecode is equal to that of a previous cluster, stop merging
• FrameVisitor to process the frames in the output of a GetMedia call.
• CopyVisitor to copy the raw bytes of the stream being parsed to an output stream.
• Add example that shows parsing and rendering Kinesis Video Streams.
• Known Issues: In KinesisVideoRendererExample, the decode/renderer sample using JCodec may not be able to decode all mkv files.

Release 1.0.2 (December 2017)

• Add example that shows integration with Kinesis Video Streams.
• Remove unnecessary import.

Release 1.0.1 (November 2017)

• Update to include the url for Amazon Kinesis Video Streams in the pom.xml

Release 1.0.0 (November 2017)

• First release of the Amazon Kinesis Video Parser Library.
• Supports Mkv elements up to version 4.
• Known issues:
  • EBMLMaxIDLength and EBMLMaxSizeLength are hardcoded as 4 and 8 bytes respectively
  • Unknown EBML elements not specified in MkvTypeInfos are not readable by the user using StreamingMkvReader.
  • Unknown EBML elements not specified in MkvTypeInfos of unknown length lead to an exception.
  • Does not do any CRC validation for any Mkv elements with the CRC-32 element.