Large scale video analytics architecture

Abstract

A video file is split into a plurality of chunks. At least a subset of the chunks is processed in parallel, including by detecting one or more moving objects and computing for each detected moving object a visual key and an associated attribute value. The visual key and the associated attribute value are provided as output.

What is claimed is: 
     
         1 . A method, comprising:
 splitting a video file into a plurality of chunks respectively comprising unstructured video data;   processing at least a subset of the chunks in parallel, including detecting one or more moving objects and computing for the one or more detected moving objects corresponding ones of a visual key and an associated-attribute value, wherein the attribute value comprises one or more of a set of one or more coordinates corresponding to a location of the moving object in an image frame with which a key-value pair is associated and a timestamp associated with the image frame with which the key-value pair is associated;   providing the visual key and the associated attribute value as an output; and   storing, in a database, the visual key and the associated attribute value as structured information in a searchable format.   
     
     
         2 . The method of  claim 1 , wherein the splitting the video file into the plurality of chunks is comprises in the event that a boundary corresponding to at least one of the plurality of chunks does not conform to a Group of Pictures boundary associated with a first Group of Pictures, sending a first subset of one or more frames corresponding to the first Group of Pictures to a processor that is to process a second subset of one or more frames corresponding to the first Group of Pictures. 
     
     
         3 . The method of  claim 1 , wherein the processing of at least the subset of chunks and computing for each detected moving object the visual key and the associated attribute value comprises:
 associating each detected moving foreground object with an image frame, a timestamp associated with the image frame, and a fingerprint representation of the corresponding moving foreground object.   
     
     
         4 . The method of  claim 1 , further comprising:
 receiving a query from a client;   querying the database based at least in part on the query;   in the event that the database includes data that is responsive to the query, providing the client with a query result, the query result including an image displaying an object corresponding to the query result; and   in response to receiving a input from the client, the input corresponding to a selection of the query result, retrieving corresponding underlying raw video data of the object corresponding to the query result.   
     
     
         5 . The method of  claim 3 , wherein the fingerprint representation of the corresponding moving foreground object is computed based on an image content analysis. 
     
     
         6 . The method of  claim 1 , wherein the one or more coordinates indicate a geo-location corresponding to the location of the moving object in the image frame. 
     
     
         7 . The method of  claim 1 , wherein the database is searchable using natural-language queries. 
     
     
         8 . The method of  claim 1 , wherein the video file comprises a standard video file format. 
     
     
         9 . The method of  claim 1 , further comprising detecting the standard video file format. 
     
     
         10 . The method of  claim 9 , further comprising transcoding the video file from the standard video file format to distributed file system-friendly format. 
     
     
         11 . The method of  claim 1 , wherein each chunk is processed at least in part by a map stage to which the chunk is provided as input. 
     
     
         12 . The method of  claim 11 , wherein the visual key and the associated attribute value are computed by the map stage and provided as output in the form of a composite key-value pair. 
     
     
         13 . The method of  claim 1 , wherein the key-value pair is sent to a reduce stage determined based at least in part on the visual key. 
     
     
         14 . The method of  claim 13 , wherein the reduce stage is configured to collect key-value pairs associated with a same moving object and to form and store a trajectory data based on the collected key-value pairs. 
     
     
         15 . The method of  claim 1 , further comprising using the visual key and associated attribute value to compute and store structured information associated with at least a portion of the video file. 
     
     
         16 . The method of  claim 15 , wherein the structured information, optional with a thumbnail image for future reference, is stored in a database. 
     
     
         17 . The method of  claim 16 , further comprising using an index associated with the structured information as stored in the database to retrieve at least a portion of said structured information from the database in response to a database query. 
     
     
         18 . The method of  claim 17 , wherein the query comprises a SQL query. 
     
     
         19 . The method of  claim 16 , wherein the database includes an advanced data analytics capability and further comprising using the advanced data analytics capability to perform advanced data analytics processing with respect to at least a portion of the structured information. 
     
     
         20 . The method of  claim 15 , further comprising providing a visual representation of at least a defined subset of said structured information. 
     
     
         21 . The method of  claim 20 , wherein the structured information comprises a trajectory of the moving object and the visual representation comprises a visual representation of the trajectory. 
     
     
         22 . The method of  claim 21 , wherein the trajectory is represented as a line or one or more other graphical elements arranged in the visual representation to indicate the trajectory of the moving object with respect to a static or other background or reference. 
     
     
         23 . The method of  claim 1 , further comprising decoding, in parallel, one or more image frames from a plurality of localized distributed chunks in a Hadoop Distributed File System (HDFS). 
     
     
         24 . A system, comprising:
 a processor; and   a memory coupled with the processor, wherein the memory is configured to provide the processor with instructions which when executed cause the processor to:
 split a video file into a plurality of chunks respectively comprising unstructured video data; 
 process at least a subset of the chunks in parallel, including detecting one or more moving objects and computing for the one or more detected moving objects corresponding ones of a visual key and an associated attribute value, wherein the attribute value comprises one or more of a set of one or more coordinates corresponding to a location of the moving object in an image frame with which a key-value pair is associated and a timestamp associated with the image frame with which the key-value pair is associated; 
 provide the visual key and the associated attribute value as an output; and 
 store, in a database, the visual key and the associated attribute value as structured information in a searchable format. 
   
     
     
         25 . A computer program product, the computer program product being embodied in a non-transitory tangible computer readable storage medium and comprising computer instructions for:
 splitting a video file into a plurality of chunks respectively comprising unstructured video data;   processing at least a subset of the chunks in parallel, including by detecting one or more moving objects and computing for the one or more detected moving objects corresponding ones of a visual key and an associated attribute value, wherein the attribute value comprises one or more of a set of one or more coordinates corresponding to a location of the moving object in an image frame with which a key-value pair is associated and a timestamp associated with the image frame with which the key-value pair is associated;   providing the visual key and the associated attribute value as an output; and   store, in a database, the visual key and the associated attribute value as structured information in a searchable format.