Anomaly detection using a kernel-based sparse reconstruction model

A system for detecting anomalies in video footage, said system comprising:

• a processor;
• a data bus coupled to said processor;
• and
• a computer-usable medium embodying computer program code, said computer-usable medium being coupled to said data bus, said computer program code comprising instructions executable by said processor and configured for:
• in an offline phase, constructing a training dictionary comprising a plurality of event classes, wherein at least one event among said plurality of event classes is defined with respect to n-dimensional feature vectors;
• in an offline phase, defining at least one nonlinear kernel function that transforms said n-dimensional feature vectors into a higher dimensional feature space;
• in an online phase, receiving at least one test event within an input video sequence of said video footage;
• and
• in an online phase, determining if said at least one test event is anomalous by deriving a sparse reconstruction with respect to said training dictionary in said higher dimensional feature space induced by said at least one nonlinear kernel function.

A processor-readable medium storing computer code representing instructions to cause a process for detecting anomalies in video footage, said computer code comprising code to:

• in an offline phase, construct a training dictionary comprising a plurality of event classes, wherein at least one event among said plurality of event classes is defined with respect to n-dimensional feature vectors;
• in an offline phase, define at least one nonlinear kernel function that transforms said n-dimensional feature vectors into a higher dimensional feature space;
• in an online phase, receive at least one test event within an input video sequence of said video footage;
• and
• in an online phase, determine if said at least one test event is anomalous by deriving a sparse reconstruction with respect to said training dictionary in said higher dimensional feature space induced by said at least one nonlinear kernel function.