Unusual event detection via collaborative video mining

A method to support mining of surveillance videos to detect abnormal event, performed by a computer system having one or more processors and memory storing one or more programs for execution by the one or more processors, the method comprising:

• segmenting one or more incoming video streams into a plurality of events;
• generating a set of abnormal events by clustering the plurality of events via unsupervised mining, including identifying clusters of the plurality of events via hierarchical clustering;
• receiving user feedback in process to identify misclassified normal events in the set of abnormal events;
• combining the set of abnormal events with the user feedback to obtain a clean training set of normal events;
• training a module for motion detection of abnormal events using the clean training set;
• and refining the motion detection result by the module via cross-validation.

A non-transitory machine readable medium storing instructions for execution by one or more processors of a computer system, the instructions comprising instructions to:

• segment one or more incoming video streams into a plurality of events;
• generate a set of abnormal events by clustering the plurality of events via unsupervised mining, including identifying clusters of the plurality of events via hierarchical clustering;
• receive user feedback in process to identify misclassified normal events in the set of abnormal events;
• combine the set of abnormal events with the user feedback to obtain a clean training set of normal events;
• train a module for motion detection of abnormal events using the clean training set;
• and refine the motion detection result by the module via cross-validation.

A system to support collaborative framework for mining of surveillance videos, comprising:

• means for segmenting one or more incoming video streams into a plurality of events;
• means for generating a set of abnormal events by clustering the plurality of events via unsupervised mining, including identifying clusters of the plurality of events via hierarchical clustering;
• means for receiving user feedback in process to identify misclassified normal events in the set of abnormal events;
• means for combining the set of abnormal events with the user feedback to obtain a clean training set of normal events;
• means for training a module for motion detection of abnormal events using the clean training set;
• and means for refining the motion detection result by the module via cross-validation.