使用 Amazon Rekognition 实时分析图像 王元恺,AWS 解决方案架构师
沟通交流 你好
沟通交流 你好
Amazon Rekognition Image 采用深度学习技术的完全托管式图像识别服务 搜索, 鉴别并组织百万的图片 对象和场景检测 面部分析面部比较面部识别名人识别图像审核文字识别
对象和场景检测 DetectLabels { }, { }, { },... "Confidence": 94.62968444824219, "Name": "adventure" "Confidence": 94.62968444824219, "Name": "boat" "Confidence": 94.62968444824219, "Name": "rafting"
对象和场景检测 用户场景
面部分析 DetectFaces [ { "BoundingBox": { "Height": 0.3449999988079071, "Left": 0.09666666388511658, "Top": 0.27166667580604553, "Width": 0.23000000417232513 }, "Confidence": 100, "Emotions": [ {"Confidence": 99.1335220336914, "Type": "HAPPY" }, {"Confidence": 3.3275485038757324, "Type": "CALM"}, {"Confidence": 0.31517744064331055, "Type": "SAD"} ], "Eyeglasses": {"Confidence": 99.8050537109375, "Value": false}, "EyesOpen": {Confidence": 99.99979400634766, "Value": true}, "Gender": {"Confidence": 100, "Value": "Female }
面部分析 Demographic Data Age Range 29-45 Gender: Male 96.5% Facial Landmarks EyeLeft,EyeRight,Nose RightPupil,LeftPupil MouthRight,LeftEyeBrowUp Bounding Box Image Quality Brightness 23.6% Sharpness 99.9% Emotions Happy 83.8% Surprised 0.65% General Attributes Smile:True 23.6% EyesOpen:True 99.8% Beard:True 99.5% Mustache:True 99.9% Facial Pose Pitch 1.446 Roll 5.725 Yaw 4.383
面部分析 用户场景 收集分析人脸图片 应用 统计和情感属性 AMAZON REKOGNITION DetectFaces 获取广告图片 日志记录 Look Your Best All Day AMAZON S3 AMAZON DYNAMODB 更新统计属性 AMAZON REDSHIFT
面孔比较 CompareFaces { } "FaceMatches": [ {"Face": {"BoundingBox": { "Height": 0.2683333456516266, "Left": 0.5099999904632568, "Top": 0.1783333271741867, "Width": 0.17888888716697693}, "Confidence": 99.99845123291016}, "Similarity": 96 }, {"Face": {"BoundingBox": { "Height": 0.2383333295583725, "Left": 0.6233333349227905, "Top": 0.3016666769981384, "Width": 0.15888889133930206}, "Confidence": 99.71249389648438}, "Similarity": 0 } ], "SourceImageFace": {"BoundingBox": { "Height": 0.23983436822891235, "Left": 0.28333333134651184, "Top": 0.351423978805542, "Width": 0.1599999964237213}, "Confidence": 99.99344635009766}
面孔比较 用户场景
transformed stored 面部识别 Face Face ID & vector<float> f7a3a278-2a59-5102-a549-a12ab1a8cae8 & v1 IndexFace 02e56305-1579-5b39-ba57-9afb0fd8782d & v2 Collection { f7a3a278-2a59-5102-a549-a12ab1a8cae8, 02e56305-1579-5b39-ba57-9afb0fd8782d, 4c55926e-69b3-5c80-8c9b-78ea01d30690 } 4c55926e-69b3-5c80-8c9b-78ea01d30690 & v3
面部识别 用户场景 相机实时信息 应用图片索引 AMAZON S3 AWS LAMBDA AMAZON REKOGNITION IndexFaces 个人信息数据表 人脸收集
名人识别 RecognizeCelebrities { "CelebrityFaces": [ { "Face": { "BoundingBox": { "Height":0.6766666769981384, "Left": 0.273333340883255, "Top": 0.09833333641290665, "Width": 0.4511111080646515 },, "Quality": { "Brightness":56.59690475463867, "Sharpness": 99.9945297241211 } }, "Id": "1SK7cR8M", "MatchConfidence": 100, "Name": "Jeff Bezos", "Urls": [ "www.imdb.com/name/nm1757263" ] } ]}
图像审核 DetectModerationLabels { "ModerationLabels": [ { Confidence : 83.55088806152344, "Name": "Suggestive", "ParentName": "" }, { Confidence :83.55088806152344, Name : Female Swimwear Or Underwear, "ParentName": "Suggestive } ] }
文字识别 DetectText { "TextDetections": [ { "Confidence": 97.4773178100586, "DetectedText": "IT'S", "Geometry": { "BoundingBox": { "Height": 0.10191240906715393, "Left": 0.6658040285110474, "Top": 0.18162749707698822, "Width": 0.15050667524337769 }, "Polygon": [ { "X": 0.6658040285110474, "Y": 0.18162749707698822 }, ] }, "Id": 0, "Type": "LINE" }, { "Confidence": 92.7912368774414, "DetectedText": "MONDAY", "Geometry": { } }
Amazon Rekognition API 总览 Non-storage API Operations Storage-based API Operations DetectModerationLabels IndexFaces SearchFacesByImage CompareFaces { "FaceMatches": [ {"Face": {"BoundingB "Height": 0.2683333456516266, "Left": 0.5099999904632568, "Top": 0.1783333271741867, "Width": 0.17888888716697693}, " DetectLabels Detect-Text CreateCollection ListCollections { "FaceMatches": [ {"Face": {"BoundingB "Height": 0.2683333456516266, "Left": 0.5099999904632568, "Top": 0.1783333271741867, "Width": 0.17888888716697693}, " SearchFaces DeleteCollection ListFaces DetectFaces RecognizeCelebrities GetCelebrityInfo DeleteFaces
Softmax Amazon Rekognition 深度学习 Labrador Probability Dog Beach Outdoors Layer 1 Layer 2 Layer n FC Feature map
Amazon Rekognition Image 客户案例 Extract the rich text captured in images at scale Processed with Amazon Rekognition for millions of Pins stored in Amazon S3 Match customers with art they will love thanks to recommendation tools built on AWS
DEMO 整体架构 前端 删除图片 实时分析 下载原图 Serverless 获取索引 图片热度 获取缩略图
DEMO 解决自定义标签 Amazon SageMaker Machine Learning AMI RiseML on Kubernetes 数据集打标签 模型训练 Custom Algorithm using existing tags
DEMO 解决单图多面部识别 Index Collection Search Delete
DEMO 实时分析图像 Amazon S3 Photos Put Event AWS Lambda URL & Height Amazon DynamoDB Presign URL Rekognition DetectFaces Rekognition IndexFaces SearchFaces DeleteFaces Rekognition DetectLabels Amazon S3 Thumbnail Initial Heat Amazon ElasticSearch Metadata
Amazon Rekognition Video 采用深度学习技术的完全托管式视频识别分析服务 对象和场景检测 轨迹跟踪 面部识别 实时流分析 不安全视频 检测 名人识别
轨迹跟踪
Amazon Rekognition Video 客户案例 Identify and track actors across millions of frames of content Allowing viewers to identify and buy the products they see on screen.
Amazon Rekognition Video 客户案例 Automating Footage Tagging with Amazon Rekognition Indexed 97,000 people Saves ~9,000 hours a year in labor
工作代码示例 https://github.com/aws-samples Search for rekognition : Smart image cropping Twitter bot Find missing persons using social media Automatic content moderation Serverless hotdog detection And more
AWS 服务集成 存储 解耦 AWS Elemental MediaConvert 视频处理 Amazon API Gateway Amazon S3 AWS Elemental MediaLive 应用 AWS Lambda Amazon EFS AWS Batch Amazon SNS Amazon EC2 Amazon Kinesis Amazon DynamoDB Amazon ECS Amazon SQS 索引 Amazon ElasticSearch 计算
Amazon Rekognition 价值 完全托管 AWS 集成经过验证的可扩展性安全低价
感谢参加 AWS INNOVATE 2018 在线技术大会 我们希望您在这里找到感兴趣的内容! 也请帮助我们完成投票打分和反馈问卷 欲获取关于 AWS 的更多信息和技术内容, 可以通过以下方式找到我们 : 微信公众号 :AWSChina 新浪微博 :https://www.weibo.com/amazonaws/ 领英 :https://www.linkedin.com/company/aws-china/ 知乎 :https://www.zhihu.com/org/aws-54/activities/ 视频中心 :http://aws.amazon.bokecc.com/ 更多线上活动 :https://aws.amazon.com/cn/about-aws/events/webinar/