System and method for abnormality detection

• US 9,852,019 B2
• Filed: 06/09/2014
• Issued: 12/26/2017
• Est. Priority Date: 07/01/2013
• Status: Active Grant

• Current Assignee: Agent Video Intelligence Ltd. (Irisity AB)
• Original Assignee: Agent Video Intelligence Ltd. (Irisity AB)
• Inventors: Ashani, Zvika
• Primary Examiner(s): Lottich, Joshua P
• Application Number:
  US14/899,276
• Publication Number:
  US 20160139977A1
• Time in Patent Office: 1,296 Days
• CPC Class Codes: G06F 11/0706 the processing taking place on a specific hardware platform or in a specific software environment
• CPC Class Codes: G06F 11/0751 Error or fault detection not based on redundancy power supply failures G06F1/30; network fault management H04L41/06
• CPC Class Codes: G06F 11/0772 Means for error signaling, e.g. using interrupts, exception flags, dedicated error registers
• CPC Class Codes: G06F 11/079 Root cause analysis, i.e. error or fault diagnosis in a hardware test environment G06F11/22; in a software test environment G06F11/36
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69

Abstract

A system and method for use in data analysis are provided. The system comprises a data processing utility configured to receive and process input data, comprising: plurality of neural network modules capable for operating in a training mode and in a data processing mode in accordance with the training; a network training utility configured for operating the neural network modules in the training mode utilizing selected set of training data pieces for sequentially training of the neural network modules in a cascade order to reduce an error value with respect to the selected set of the training data pieces for each successive neural network module in the cascade; and an abnormality detection utility configured for sequentially operating said neural network modules for processing input data, and classifying said input data as abnormal upon identifying that all the neural network modules provide error values being above corresponding abnormality detection thresholds.

Claims

1. A method for use in abnormality detection, the method comprising:

providing a plurality of neural network modules; and

sequentially training said neural network modules in a cascade order with respect to a set of training data pieces to thereby reduce an error value for each successive neural network module, thereby minimizing an error value with respect to said training data pieces for each successive neural network module in the cascade;

wherein said sequentially training comprising;

a. providing a first set of calibration data pieces and generating therefrom a first normal behavior map corresponding to said first set of calibration data,

b. determining a second set of calibration data pieces which includes calibration data pieces of said first set of calibration data pieces being outside of said first normal behavior map,

c. using said second set of calibration data pieces and generating a corresponding second normal behavior map corresponding to said second set of calibration data,

d. determining a third set of calibration data pieces including calibration data pieces being outside of said second normal behavior map,

e. repeating the process until number of calibration data pieces in a newly generated set satisfies at least one of the following;

the number equals to number of calibration data pieces in previous set, or the number is below a predetermined threshold.

2. The method of claim 1, comprising producing for each set of calibration data pieces a corresponding set of input vectors and scaling said set of input vectors with respect to mean and standard deviation values of at least one parameter of said input vectors.

3. The method of claim 1, comprising:

receiving input data pieces in which abnormalities are to be detected, comparing said input data pieces to the first normal behavior map and classifying said input data pieces as either normal or abnormal with respect to said first behavior map, and electively applying a comparison step to the abnormal data pieces, said selective comparison step comprising comparing the abnormal input data pieces to the second behavior map to further classify said abnormal input data pieces as either normal or abnormal with respect to the second behavior map, and repeating the selective comparison step until all the input data pieces are classified as normal, and upon identifying that there exist at least one input data piece classified as abnormal with respect to all behavior maps, generating data indicative of said at least one abnormal input data piece.

4. The method of claim 1, wherein said generating of a normal behavior map comprising utilizing a neural network based on a set of data pieces.

5. The method of claim 1, wherein said data pieces comprise object related data pieces, generated from an image stream.