Amazon Rekognition
Developer Guide

People Pathing

Amazon Rekognition Video can create a track of the path people take in videos and provide information such as:

  • The location of the person in the video frame at the time their path is tracked.

  • Facial landmarks such as the position of the left eye, when detected.

Amazon Rekognition Video people pathing in stored videos is an asynchronous operation. To start the pathing of people in videos call StartPersonTracking. Amazon Rekognition Video publishes the completion status of the video analysis to an Amazon Simple Notification Service topic. If the video analysis is succesful, call GetPersonTracking to get results of the video analysis. For more information about calling Amazon Rekognition Video API operations, see Calling Amazon Rekognition Video Operations.

The following procedure shows how to track the path of people through a video stored in an Amazon S3 bucket. The example expands on the code in Analyzing a Video Stored in an Amazon S3 Bucket with Java or Python (SDK) which uses an Amazon Simple Queue Service queue to get the completion status of a video analysis request.

To detect people in a video stored in an Amazon S3 bucket (SDK)

  1. Perform Analyzing a Video Stored in an Amazon S3 Bucket with Java or Python (SDK).

  2. Add the following code to the class VideoDetect that you created in step 1.

    //Copyright 2018, Inc. or its affiliates. All Rights Reserved. //PDX-License-Identifier: MIT-0 (For details, see //Persons======================================================================== private static void StartPersons(String bucket, String video) throws Exception{ int maxResults=10; String paginationToken=null; StartPersonTrackingRequest req = new StartPersonTrackingRequest() .withVideo(new Video() .withS3Object(new S3Object() .withBucket(bucket) .withName(video))) .withNotificationChannel(channel); StartPersonTrackingResult startPersonDetectionResult = rek.startPersonTracking(req); startJobId=startPersonDetectionResult.getJobId(); } private static void GetResultsPersons() throws Exception{ int maxResults=10; String paginationToken=null; GetPersonTrackingResult personTrackingResult=null; do{ if (personTrackingResult !=null){ paginationToken = personTrackingResult.getNextToken(); } personTrackingResult = rek.getPersonTracking(new GetPersonTrackingRequest() .withJobId(startJobId) .withNextToken(paginationToken) .withSortBy(PersonTrackingSortBy.TIMESTAMP) .withMaxResults(maxResults)); VideoMetadata videoMetaData=personTrackingResult.getVideoMetadata(); System.out.println("Format: " + videoMetaData.getFormat()); System.out.println("Codec: " + videoMetaData.getCodec()); System.out.println("Duration: " + videoMetaData.getDurationMillis()); System.out.println("FrameRate: " + videoMetaData.getFrameRate()); //Show persons, confidence and detection times List<PersonDetection> detectedPersons= personTrackingResult.getPersons(); for (PersonDetection detectedPerson: detectedPersons) { long seconds=detectedPerson.getTimestamp()/1000; System.out.print("Sec: " + Long.toString(seconds) + " "); System.out.println("Person Identifier: " + detectedPerson.getPerson().getIndex()); System.out.println(); } } while (personTrackingResult !=null && personTrackingResult.getNextToken() != null); }

    2a. In the function main, replace the line:




    2b. Replace the line:




    #Copyright 2018, Inc. or its affiliates. All Rights Reserved. #PDX-License-Identifier: MIT-0 (For details, see def GetResultsPersons(self, jobId): maxResults = 10 paginationToken = '' finished = False while finished == False: response = self.rek.get_person_tracking(JobId=jobId, MaxResults=maxResults, NextToken=paginationToken) print(response['VideoMetadata']['Codec']) print(str(response['VideoMetadata']['DurationMillis'])) print(response['VideoMetadata']['Format']) print(response['VideoMetadata']['FrameRate']) for personDetection in response['Persons']: print('Index: ' + str(personDetection['Person']['Index'])) print('Timestamp: ' + str(personDetection['Timestamp'])) print() if 'NextToken' in response: paginationToken = response['NextToken'] else: finished = True

    2a. In the function main, replace the line:

    response = self.rek.start_label_detection(Video={'S3Object': {'Bucket': self.bucket, 'Name':}}, NotificationChannel={'RoleArn': self.roleArn, 'SNSTopicArn': self.topicArn})


    response = self.rek.start_person_tracking(Video={'S3Object':{'Bucket':self.bucket,'Name'}}, NotificationChannel={'RoleArn':self.roleArn, 'SNSTopicArn':self.topicArn})

    2b. Replace the line:





    If you've already run a video example other than Analyzing a Video Stored in an Amazon S3 Bucket with Java or Python (SDK), the function name to replace is different.

  3. Run the code. The unique identifiers for tracked people are shown along with the time, in seconds, the people's paths were tracked.

GetPersonTracking Operation Response

GetPersonTracking returns an array, Persons, of PersonDetection objects which contain details about people detected in the video and when their paths are tracked.

You can sort Persons by using the SortBy input parameter. Specify TIMESTAMP to sort the elements by the time people's paths are tracked in the video. Specify INDEX to sort by people tracked in the video. Within each set of results for a person, the elements are sorted by descending confidence in the accuracy of the path tracking. By default, Persons is returned sorted by TIMESTAMP. The following example is the JSON response from GetPersonDetection. The results are sorted by the time, in milliseconds since the start of the video, that people's paths are tracked in the video. In the response, note the following:

  • Person information – The PersonDetection array element contains information about the detected person. For example, the time the person was detected (Timestamp), the position of the person in the video frame at the time they were detected (BoundingBox), and how confident Amazon Rekognition Video is that the person has been correctly detected (Confidence).

    Facial features are not returned at every timestamp for which the person's path is tracked. Furthermore, in some circumstances a tracked person's body might not be visible, in which case only their face location is returned.

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

  • Index – A unique identifier for identifying the person throughout the video.

  • Video information – The response includes information about the video format (VideoMetadata) in each page of information returned by GetPersonDetection.

{ "JobStatus": "SUCCEEDED", "NextToken": "AcDymG0fSSoaI6+BBYpka5wVlqttysSPP8VvWcujMDluj1QpFo/vf+mrMoqBGk8eUEiFlllR6g==", "Persons": [ { "Person": { "BoundingBox": { "Height": 0.8787037134170532, "Left": 0.00572916679084301, "Top": 0.12129629403352737, "Width": 0.21666666865348816 }, "Face": { "BoundingBox": { "Height": 0.20000000298023224, "Left": 0.029999999329447746, "Top": 0.2199999988079071, "Width": 0.11249999701976776 }, "Confidence": 99.85971069335938, "Landmarks": [ { "Type": "eyeLeft", "X": 0.06842322647571564, "Y": 0.3010137975215912 }, { "Type": "eyeRight", "X": 0.10543643683195114, "Y": 0.29697132110595703 }, { "Type": "nose", "X": 0.09569807350635529, "Y": 0.33701086044311523 }, { "Type": "mouthLeft", "X": 0.0732642263174057, "Y": 0.3757539987564087 }, { "Type": "mouthRight", "X": 0.10589495301246643, "Y": 0.3722417950630188 } ], "Pose": { "Pitch": -0.5589138865470886, "Roll": -5.1093974113464355, "Yaw": 18.69594955444336 }, "Quality": { "Brightness": 43.052337646484375, "Sharpness": 99.68138885498047 } }, "Index": 0 }, "Timestamp": 0 }, { "Person": { "BoundingBox": { "Height": 0.9074074029922485, "Left": 0.24791666865348816, "Top": 0.09259258955717087, "Width": 0.375 }, "Face": { "BoundingBox": { "Height": 0.23000000417232513, "Left": 0.42500001192092896, "Top": 0.16333332657814026, "Width": 0.12937499582767487 }, "Confidence": 99.97504425048828, "Landmarks": [ { "Type": "eyeLeft", "X": 0.46415066719055176, "Y": 0.2572723925113678 }, { "Type": "eyeRight", "X": 0.5068183541297913, "Y": 0.23705792427062988 }, { "Type": "nose", "X": 0.49765899777412415, "Y": 0.28383663296699524 }, { "Type": "mouthLeft", "X": 0.487221896648407, "Y": 0.3452930748462677 }, { "Type": "mouthRight", "X": 0.5142884850502014, "Y": 0.33167609572410583 } ], "Pose": { "Pitch": 15.966927528381348, "Roll": -15.547388076782227, "Yaw": 11.34195613861084 }, "Quality": { "Brightness": 44.80223083496094, "Sharpness": 99.95819854736328 } }, "Index": 1 }, "Timestamp": 0 }..... ], "VideoMetadata": { "Codec": "h264", "DurationMillis": 67301, "FileExtension": "mp4", "Format": "QuickTime / MOV", "FrameHeight": 1080, "FrameRate": 29.970029830932617, "FrameWidth": 1920 } }