Intelligent Video Recognition
MPS leverages AI technologies to recognize video content. The result of an intelligent video recognition task includes a recognition score, suggestion, and suspicious video segments. You can decide whether to expose a video based on the suggestion.
Result
MPS can intelligently recognize video images, speech (ASR), and optical characters (OCR).
Object | Operation | Description |
Video images
(People and objects) | Pornographic content | Checks for pornographic content in video images, including:`vulgar`: vulgarity`intimacy`: intimacy`sexy`: sexiness |
| Politically sensitive content | Checks for politically sensitive content in video images, including:`bloody`: bloodiness`explosion`: explosions and fires`violation_photo`: banned icons`guns`: weapons and guns |
Speech
(Speech to text) | Pornographic content | Checks for keywords for pornographic content in speech |
| Politically sensitive content | Checks for keywords for politically sensitive content in speech |
Optical characters
Image to text | Pornographic content | Checks for keywords for pornographic content in images |
| Politically sensitive content | Checks for keywords for politically sensitive content in images |
Parameter description
Field Name | Type | Description |
confidence | Float | Intelligent recognition score (0-100). The higher the score, the more suspicious the content. |
suggestion | String | There are three types of intelligent recognition suggestions: pass : The degree of suspicion is not high, and approval is recommended. review : The degree of suspicion is high, and human review is recommended. block : The degree of suspicion is very high, and blocking is recommended. |
segments | Array | Suspicious video segments, which help you locate suspicious content in a video |
Initiating Task
Directions
You can call an API to initiate an intelligent video recognition task or configure automatic triggering of the task upon video upload.
API: Call the ProcessMedia API, setting
AiContentReviewTask
to the ID of your intelligent video recognition template.Automatic triggering upon upload: In the console, create a workflow with intelligent video recognition enabled and upload videos to the bucket bound to the workflow.
Creating template
MPS uses templates to represent combinations of intelligent video recognition parameters, which determine which of the following operations MPS performs.
Recognition of pornographic content in video images
Recognition of politically sensitive content in video images
Recognition of pornographic keywords in speech (ASR)
Recognition of politically sensitive keywords in speech (ASR)
Recognition of pornographic keywords in images (OCR)
Recognition of politically sensitive keywords in images (OCR)
MPS provides preset intelligent video recognition templates for common parameter combinations. You can also use a server API to create and manage custom templates.
Obtaining Result
After initiating an intelligent video recognition task, you can wait for the result notification asynchronously or query the result synchronously.
Below is an example of the result returned after query (fields with null values are omitted):
{"TaskType":"WorkflowTask","Status":"FINISH","CreateTime":"2019-07-16T06:21:27Z","BeginProcessTime":"2019-07-16T06:21:28Z","FinishTime":"2019-07-16T06:21:46Z","WorkflowTask":{"TaskId":"2356768367-WorkflowTask-2e1af2456351812be963e309cc133403t0","Status":"FINISH","InputInfo":{"Type":"COS","CosInputInfo":{"Bucket":"MyVideoBucket-235303****","Region":"ap-beijing","Object":"/input/AnimalWorld.mp4"}},"MetaData":{"AudioDuration":60,"AudioStreamSet":[{"Bitrate":383854,"Codec":"aac","SamplingRate":48000}],"Bitrate":1021028,"Container":"mov,mp4,m4a,3gp,3g2,mj2","Duration":60,"Height":480,"Rotate":0,"Size":7700180,"VideoDuration":60,"VideoStreamSet":[{"Bitrate":637174,"Codec":"h264","Fps":23,"Height":480,"Width":640}],"Width":640},"MediaProcessResultSet":[],"AiContentReviewResultSet":[{"Type":"Porn","PornTask":{"Status":"SUCCESS","ErrCode":0,"Message":"","Input":{"Definition":10},"Output":{"Confidence":98,"Suggestion":"block","Label":"sexy","SegmentSet":[{"StartTimeOffset":9.5,"EndTimeOffset":14,"Confidence":98,"Suggestion":"block","Label":"sexy","Url":"http://xxx.vod2.myqcloud.com/xxx/xxx/xx1.jpg","PicUrlExpireTime":"2019-07-23T06:21:46Z"},{"StartTimeOffset":16.5,"EndTimeOffset":18,"Confidence":80,"Suggestion":"review","Label":"sexy","Url":"http://xxx.vod2.myqcloud.com/xxx/xxx/xx2.jpg","PicUrlExpireTime":"2019-07-23T06:21:46Z"},{"StartTimeOffset":41,"EndTimeOffset":49,"Confidence":97,"Suggestion":"block","Label":"sexy","Url":"http://xxx.vod2.myqcloud.com/xxx/xxx/xx3.jpg","PicUrlExpireTime":"2019-07-23T06:21:46Z"}]}}},{"Type":"Terrorism","TerrorismTask":{"Status":"SUCCESS","ErrCode":0,"Message":"","Input":{"Definition":10},"Output":{"Confidence":0,"Suggestion":"pass","SegmentSet":[]}}},{"Type":"Political","PoliticalTask":{"Status":"SUCCESS","ErrCode":0,"Message":"","Input":{"Definition":10},"Output":{"Confidence":0,"Suggestion":"pass","SegmentSet":[]}}}],"AiAnalysisResultSet":[],"AiRecognitionResultSet":[]},"TasksPriority":0,"SessionId":"","SessionContext":"","RequestId":"xxx-xxx-xxx"}
As shown above, there are three types of results under
WorkflowTask.AiContentReviewResultSet
: Porn
, Terrorism
, and Political
.For
Porn
, Output.Suggestion
is block
, which indicates a very high likelihood that the content is pornographic, and you are advised to block it. The confidence score is 98, and the label for the content is sexy
.Three suspicious video segments are identified for
Porn
, whose start and end times are specified by StartTimeOffset
and EndTimeOffset
.According to the results for
Terrorism
and Political
, no inappropriate content is detected in the video.