Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
  • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
  • G06F16/90—Details of database functions independent of the retrieved data types
  • G06F16/903—Querying
  • G06F16/90335—Query processing

Definitions

• the invention relates to a method for optimizing the search for a scene from an image stream archived in a video database.
• the invention also relates to software stored on a recording medium and intended, when executed by a computer, to implement the method, and a device adapted to implement the method.
• Video images are increasingly used in industrial applications for process control and monitoring and in video surveillance applications of public or private sites. These applications typically use a network of cameras wisely arranged to provide reliable images at different points in the monitored space. The images provided by the different cameras are compressed and then stored in a video database for later use. In most applications, especially in the field of video surveillance, this operation requires the processing of a large volume of video images, especially when the network has a large number of cameras scattered in a large area such as a city for example. The quantity of stored images quickly becomes too important for an operator to perform a fast and efficient analysis of the images in order to extract facts or objects relevant to the intended application.
• an investigation may require the viewing and / or processing of several tens of thousands of hours of video. It is then difficult to find the information sought if no prior indexing of the videos was undertaken as soon as the acquisition. Moreover, the videos available during the investigation are the ones that have been stored, thus compressed and no longer have the optimal image quality for a richest possible information extraction.
• events and indications that generate alarms may be insufficient to navigate the archives quickly and efficiently in search of objects (individuals, vehicles) that may provide relevant information. This is the case, for example, of finding suspicious individuals in a crowd at different points in an area monitored by cameras.
• a disadvantage of current CCTV systems is that they focus exclusively on current events and generate alarms for predefined events.
• the notion of "memory" of such a system is limited to recorded videos and detected alarms.
• the systems do not allow to find an event that did not generate an alarm when it occurred, but which became determinant in the context of a subsequent investigation.
• a first object of the invention is to organize the memory of such systems so that they allow an effective investigation by limiting the amount of data to be analyzed by the operator, and by systematically annotating the streams of images obtained by the cameras to allow a quick selection of video sequences relevant to the investigation.
• a second object of the invention and to provide hardware and software tools enabling the operator to navigate quickly and efficiently in video archives by means of systematic indexing to extract the information on the stream before compression to benefit from a maximum image quality.
• the invention provides a method of assisting the investigation into a video archive based on the generic and systematic annotation of the streams, on the filtering of uninteresting videos from generic queries, and on the selection of the relevant videos. from a targeted search.
• the process according to the invention comprises: a first pretreatment phase, prior to archiving said images, comprising the following steps:
• a third search phase among the pre-selected video segments of a particular scene from at least one additional feature not forming part of the annotations associated with the preselected video segments.
• said third phase comprises a learning step allowing the recognition of said additional characteristic in the preselected video segments.
• the third search phase of a particular scene is executed by means of a generic request including said additional feature and the annotations associated with the preselected video segments.
• additional information retrieval processes are applied. on the pre-selected segments, and said additional information is compared with information contained in the learning models of the specific characteristic sought.
• the generic information extracted from said images is defined according to the intended application of the desired scene.
• a possible application of the method according to the invention relates to the search for a particular scene in a stream of images obtained by a network of CCTV cameras.
• the annotation of the scenes of the stream of images obtained by the cameras is carried out independently, flow by stream, on each of the streams obtained by each camera of the CCTV network.
• the annotation of said scenes is performed by processing the annotations associated with several distinct streams, either by the preprocessing unit or by the processing unit.
• the annotation of said scenes can be achieved by merging the annotations associated with several distinct streams.
• This embodiment is particularly suitable for a video surveillance application made by a system comprising several cameras for example.
• the first pretreatment phase is performed during the acquisition of the images.
• the first preprocessing phase is performed during the archiving of the images.
• the method is implemented by software stored on a recording medium and capable of being executed by a computer.
• This software includes:
• a first module comprising instructions for carrying out a pretreatment phase prior to the archiving of said images comprising the following steps: extracting generic information from the images of said stream,
• a second module comprising instructions for carrying out an investigation phase by preselecting video segments in the image stream comprising annotations indications associated with said images, and,
• a third module comprising instructions for carrying out a search phase of a particular scene from at least one additional feature that is not part of annotations associated with the preselected video segments.
• Said software is capable of being implemented in a device for optimizing the search for a scene from an image stream archived in a video database comprising:
• a first unit intended to perform a pretreatment of the images of the stream prior to the archiving of said images, said first unit comprising:
• a third unit intended to perform a search phase of a particular scene from at least one additional feature not forming part of the annotations associated with the preselected video segments.
• FIG. 1 schematically illustrates a general architecture of an acquisition system and of images in which is implemented the method according to the invention
• - Figure 2 shows a block diagram illustrating the essential steps of video segment search according to the invention.
• FIG. 1 schematically illustrates a system for acquiring and processing images obtained by a network of cameras 2 in a video surveillance network for example.
• the output of each camera 2 is connected to a first preprocessing unit 4 comprising an annotation module 6 and a compression module 8.
• the output of each pretreatment unit 4 is connected to a second preprocessing unit 10 comprising a module annotation merging 12 and a memory 14.
• the output of the second pretreatment unit 10 is connected to an indexing unit 16 which can be central or distributed and communicating with a search unit 18.
• the search unit 18 is also related to a postprocessing module 19 which supports learning and searching for the specific characteristic sought.
• the images obtained by a camera 2 are transmitted, before compression, to the first preprocessing unit 4 associated with it.
• the annotation module 6 of the processing unit 4 comprises software whose functions are adaptable to the application envisaged for adding generic annotations to the images received from the camera.
• a video surveillance application it may be to detect and characterize moving objects (pedestrians, vehicles).
• Generic annotations are for example "pedestrians", “vehicles”, object-related trajectories, characterization attributes, etc. In an industrial process monitoring application, this may be the detection and characterization of objects parading on a treadmill.
• the generic annotations are for example the shape or the color of the objects.
• the annotation of the scenes is done by processing the annotations associated with several distinct streams. It can be done independently by stream or by merging the annotations made on each stream, by means of the annotation merge module 12, annotations of several streams (multi-camera tracking video surveillance for example).
• the annotations can be extracted locally as close as possible to the cameras (within the preprocessing unit 4) or before indexing in the processing unit 12.
• the stored images then undergo systematic indexing in the indexing unit 16.
• This indexing makes it possible to quickly eliminate the images that do not correspond to the search criteria specified by the operator during the investigation phase.
• we are interested in a pedestrian in a video archive all the sequences presenting only cars will be automatically eliminated from the search.
• the navigation in the database can be optimized by a non-generic characteristics recognition learning phase applied to the software of the preprocessing unit 4.
• the post-processing unit 19 can be configured to automatically select sequences of pedestrians carrying a red backpack among preselected segments from generic features (ie pedestrian sequences).
• this search is preceded by a step of indexing and preselecting the shots containing human beings, etc.
• Figure 2 schematically illustrates the steps of searching a video segment from a database.
• step 20 the operator initiates a search request of the video segment by means of the search unit 18.
• This request essentially comprises generic criteria associated with the object sought during the preprocessing phase.
• step 22 the indexing unit 16 searches for the object or objects corresponding to said generic criteria among the preselected segments during the preprocessing phase and transmits the segments found to the post-processing unit 19
• the search is then optimized by means of a model obtained from a modeling unit 26 included in the post-processing unit 19.
• the modeling unit 26 constructs models taking into account generic annotations and additional specific criteria.
• the invention can be implemented in intelligent video surveillance systems that can be used in the context of investigations and in a more general way for all targeted research in a database of videos (industrial vision, multimedia, ).
• the method makes it possible to considerably reduce the number of necessary operators and the search times for particular individuals, vehicles or events.

Applications Claiming Priority (2)

Publications (1)

Family

ID=40521950

Family Applications (1)

Country Status (6)

Families Citing this family (4)

Family Cites Families (6)

• 2008
  • 2008-09-30 FR FR0856581A patent/FR2936627B1/fr not_active Expired - Fee Related
• 2009
  • 2009-09-28 WO PCT/EP2009/062507 patent/WO2010037704A1/fr active Application Filing
  • 2009-09-28 JP JP2011528355A patent/JP5548202B2/ja not_active Expired - Fee Related
  • 2009-09-28 US US13/120,340 patent/US9275140B2/en active Active
  • 2009-09-28 EP EP09783466A patent/EP2332079A1/fr not_active Withdrawn
• 2011
  • 2011-03-16 IL IL211768A patent/IL211768A/en not_active IP Right Cessation

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events