aws-samples / amazon-rekognition-automate-video-editing-example Goto Github PK

For a whole video contaning many actors/speakers, we can just pull out snippets and create a video with those snippets so that it contains content from a single speaker/actor.

In this example, we are using this video: What is it like to be a Solution Architect and then we are extracting Lia's bit(snippets where she is present) by using her image. We have uploaded both(the video and Lia's image) in an s3 bucket and then use Amazon Rekognition to perform video editing.

1. Have aws-cli installed

This is done in below Steps:

1. Create an s3 bucket manually and give a name. eg we are naming it : my-rekognition-images

2. Add the video and referenced image that need to be searched in the video in the above created bucket. For this example I am uploading a video SAs.mp4 and Lia.JPG

3. Create an Elastic Transcoder Pipeline and provide the above s3 bucket as shown below:

4. Note the pipeline id by clicking the 'search icon' as shown below

5. Set up the infrastructure by launching the template in CloudFormation in your aws account and specify the pieline-id you got from step 3.

   • Note the SNSArn and RekognitionSNSPublishRoleArn from the Output section in the CloudFormation stack

6. Run the below commands in aws-cli

   a. Create a collection

   aws rekognition create-collection --collection-id my-collection --region ap-southeast-2

   b. Add image/face to be indexed

   aws rekognition index-faces --collection-id my-collection --image "S3Object={Bucket=my-rekognition-images,Name=Lia.JPG}" --external-image -id Lia --region ap-southeast-2

   c. Start the face search in the video

   aws rekognition start-face-search --video "S3Object={Bucket=my-rekognition-images,Name=SAs.mp4}" --collection-id my-collection --notification-channel '{\"SNSTopicArn\":\"<sns-topic-arn-found-at-step-5>\", \"RoleArn\":\"<role-arn-found-at-step-5>\"}' --region ap-southeast-2

   This is an asyncronous process and once the job is done, Amazon Rekognition will trigger the SNS topic which in turn triggers the lambda function which will do the job for us and the output video output.mp4 will be uploaded in the my-rekognition-images s3 bucket

See CONTRIBUTING for more information.

This library is licensed under the MIT-0 License. See the LICENSE file.