Using the Amazon Rekognition Segment API - Amazon Rekognition

Using the Amazon Rekognition Segment API

Amazon Rekognition Video segment detection in stored videos is an Amazon Rekognition Video asynchronous operation. The Amazon Rekognition Segment API is a composite API where you choose the type of analysis (technical cues or shot detection) from a single API call. For information about calling asynchronous operations, see Calling Amazon Rekognition Video operations.

Starting segment analysis

To start the detection of segments in a stored video call StartSegmentDetection. The input parameters are the same as other Amazon Rekognition Video operations with the addition of segment type selection and result filtering. For more information, see Starting video analysis.

The following is example JSON passed by StartSegmentDetection. The request specifies that both technical cue and shot detection segments are detected. Different filters for the minimum detection confidence are requested for technical cue segments (90%) and shot detection segments (80%).

{ "Video": { "S3Object": { "Bucket": "test_files", "Name": "test_file.mp4" } "SegmentTypes":["TECHNICAL_CUES", "SHOT"] "Filters": { "TechnicalCueFilter": { "MinSegmentConfidence": 90, "BlackFrame" : { "MaxPixelThreshold": 0.1, "MinCoveragePercentage": 95 } }, "ShotFilter" : { "MinSegmentConfidence": 60 } } }

Choosing a segment type

Use the SegmentTypes array input parameter to detect technical cue and/or shot detection segments in the input video.

  • TECHNICAL_CUE — identifies frame-accurate timestamps for the start, end, and duration of technical cues (black frames, color bars, opening credits, end credits, studio logos, and primary program content) detected in a video. For example, you can use technical cues to find the start of the end credits. For more information, see Technical cues.

  • SHOT — Identifies the start, end, and duration of a shot. For example, you can use shot detection to identify candidate shots for a final edit of a video. For more information, see Shot detection.

Filtering the analysis results

You can use the Filters (StartSegmentDetectionFilters) input parameter to specify the minimum detection confidence returned in the response. Within Filters, use ShotFilter (StartShotDetectionFilter) to filter detected shots. Use TechnicalCueFilter (StartTechnicalCueDetectionFilter) to filter technical cues.

For example code, see Example: Detecting segments in a stored video.

Getting segment analysis results

Amazon Rekognition Video publishes the completion status of the video analysis to an Amazon Simple Notification Service topic. If the video analysis is successful, call GetSegmentDetection to get results of the video analysis.

The following is an example GetSegmentDetection request. The JobId is the job identifier returned from the call to StartSegmentDetection. For information about the other input parameters, see Getting Amazon Rekognition Video analysis results.

{ "JobId": "270c1cc5e1d0ea2fbc59d97cb69a72a5495da75851976b14a1784ca90fc180e3", "MaxResults": 10, "NextToken": "XfXnZKiyMOGDhzBzYUhS5puM+g1IgezqFeYpv/H/+5noP/LmM57FitUAwSQ5D6G4AB/PNwolrw==" }

GetSegmentDetection returns results for the requested analysis and general information about the stored video.

General information

GetSegmentDection returns the following general information.

  • Audio information — The response includes audio metadata in an array, AudioMetadata, of AudioMetadata objects. There can be multiple audio streams. Each AudioMetadata object contains metadata for a single audio stream. Audio information in an AudioMetadata objects includes the audio codec, the number of audio channels, the duration of the audio stream, and the sample rate. Audio metadata is returned in each page of information returned by GetSegmentDetection.

  • Video information – Currently, Amazon Rekognition Video returns a single VideoMetadata object in the VideoMetadata array. The object contains information about the video stream in the input file that Amazon Rekognition Video chose to analyze. The VideoMetadata object includes the video codec, video format and other information. Video metadata is returned in each page of information returned by GetSegmentDetection.

  • Paging information – The example shows one page of segment information. You can specify how many elements to return in the MaxResults input parameter for GetSegmentDetection. If more results than MaxResults exist, GetSegmentDetection returns a token (NextToken) used to get the next page of results. For more information, see Getting Amazon Rekognition Video analysis results.

  • Request information – The type of analysis requested in the call to StartSegmentDetection is returned in the SelectedSegmentTypes field.


Technical cues and shot information detected in a video is returned in an array, Segments, of SegmentDetection objects. The array is sorted by the segment types (TECHNICAL_CUE or SHOT) specified in the SegmentTypes input parameter of StartSegmentDetection. Within each segment type the array is sorted by timestamp values. Each SegmentDetection object includes information about the type of detected segment (Technical cue or shot detection) and general information, such as the start time, end time, and the duration of the segment.

Time information is returned in three formats.

  • Milliseconds

    The number of milliseconds since the start of the video. The fields DurationMillis, StartTimestampMillis, and EndTimestampMillis are in millisecond format.

  • Timecode

    Amazon Rekognition Video timecodes are in SMPTE format where each frame of video has a unique timecode value. The format is hh:mm:ss:frame. For example, a timecode value of 01:05:40:07, would be read as one hour, five minutes, forty seconds and seven frames. Drop frame rate use cases are supported by Amazon Rekognition Video. The drop rate timecode format is hh:mm:ss;frame. The fields DurationSMPTE, StartTimecodeSMPTE, and EndTimecodeSMPTE are in timecode format.

  • Frame Counters

    The duration of each video segment is also expressed with the number of frames. The field StartFrameNumber gives the frame number at the start of a video segment, and EndFrameNumber gives the frame number at the end of a video segment. DurationFrames gives the total number of frames in a video segment. These values are calculated using a frame index that starts with 0.

You can use the SegmentType field to determine the type of a segment returned by Amazon Rekognition Video.

  • Technical Cues – the TechnicalCueSegment field is an TechnicalCueSegment object that contains the detection confidence and the type of a technical cue. The types of technical cue are ColorBars, EndCredits, BlackFrames, OpeningCredits, StudioLogo, Slate, and Content.

  • Shot – the ShotSegment field is a ShotSegment object than contains the detection confidence and an identifier for the shot segment within the video.

The following example is the JSON response from GetSegmentDetection.

{ "SelectedSegmentTypes": [ { "ModelVersion": "2.0", "Type": "SHOT" }, { "ModelVersion": "2.0", "Type": "TECHNICAL_CUE" } ], "Segments": [ { "DurationFrames": 299, "DurationSMPTE": "00:00:09;29", "StartFrameNumber": 0, "EndFrameNumber": 299, "EndTimecodeSMPTE": "00:00:09;29", "EndTimestampMillis": 9976, "StartTimestampMillis": 0, "DurationMillis": 9976, "StartTimecodeSMPTE": "00:00:00;00", "Type": "TECHNICAL_CUE", "TechnicalCueSegment": { "Confidence": 90.45006561279297, "Type": "BlackFrames" } }, { "DurationFrames": 150, "DurationSMPTE": "00:00:05;00", "StartFrameNumber": 299, "EndFrameNumber": 449, "EndTimecodeSMPTE": "00:00:14;29", "EndTimestampMillis": 14981, "StartTimestampMillis": 9976, "DurationMillis": 5005, "StartTimecodeSMPTE": "00:00:09;29", "Type": "TECHNICAL_CUE", "TechnicalCueSegment": { "Confidence": 100.0, "Type": "Content" } }, { "DurationFrames": 299, "ShotSegment": { "Index": 0, "Confidence": 99.9982681274414 }, "DurationSMPTE": "00:00:09;29", "StartFrameNumber": 0, "EndFrameNumber": 299, "EndTimecodeSMPTE": "00:00:09;29", "EndTimestampMillis": 9976, "StartTimestampMillis": 0, "DurationMillis": 9976, "StartTimecodeSMPTE": "00:00:00;00", "Type": "SHOT" }, { "DurationFrames": 149, "ShotSegment": { "Index": 1, "Confidence": 99.9982681274414 }, "DurationSMPTE": "00:00:04;29", "StartFrameNumber": 300, "EndFrameNumber": 449, "EndTimecodeSMPTE": "00:00:14;29", "EndTimestampMillis": 14981, "StartTimestampMillis": 10010, "DurationMillis": 4971, "StartTimecodeSMPTE": "00:00:10;00", "Type": "SHOT" } ], "JobStatus": "SUCCEEDED", "VideoMetadata": [ { "Format": "QuickTime / MOV", "FrameRate": 29.970029830932617, "Codec": "h264", "DurationMillis": 15015, "FrameHeight": 1080, "FrameWidth": 1920, "ColorRange": "LIMITED" } ], "AudioMetadata": [ { "NumberOfChannels": 1, "SampleRate": 48000, "Codec": "aac", "DurationMillis": 15007 } ] }

For example code, see Example: Detecting segments in a stored video.