US 12,392,860 B2
System and method for event recognition
Jian Zhang, Denistone East (AU); Yi Da Xu, Broadway (AU); and Zhenguo Shi, Sydney (AU)
Assigned to IQ WORKS LIMITED, Chai Wan (HK)
Appl. No. 17/639,693
Filed by IQ Works Limited, Chai Wan (HK)
PCT Filed Sep. 1, 2020, PCT No. PCT/CN2020/112825
§ 371(c)(1), (2) Date Mar. 2, 2022,
PCT Pub. No. WO2021/043126, PCT Pub. Date Mar. 11, 2021.
Claims priority of application No. 19129029.5 (HK), filed on Sep. 2, 2019.
Prior Publication US 2022/0291326 A1, Sep. 15, 2022
Int. Cl. G01S 5/02 (2010.01); G06N 20/00 (2019.01)
CPC G01S 5/02521 (2020.05) [G06N 20/00 (2019.01)] 28 Claims
OG exemplary drawing
 
1. A method for event recognition of one or more target activities in a monitored space, the method comprising the steps of:
receiving and reading by a signal reader, disposed in the monitored space, an electromagnetic frequency (EMF) signal transmitted by a signal emitter over a defined space, wherein the signal includes channel state information (CSI) that is affected by an event of any persons, animals or objects between the signal emitter and the signal reader within the defined space,
pre-processing by a pre-processing processor for extracting activities related information and constructing a correlation feature matrix (CFM), wherein the step of extracting activities related information comprises the step of reducing the CSI for static background objects by subtracting static objects CSI constructed from a linear recursive operation;
processing the CFM with a matching engine, wherein the engine is arranged to train and extract with a learning network hidden features from the CFM of the signal for classifying the signal with one or more labels associated with the event, wherein the learning network is machine trained with a training data set having a plurality of CFM having amplitude and phase matrices of the signal and at least one output label associated with the event that affected the CSI of the signal, and determining whether the one or more labels are associated with a target activity; and
outputting a notification to a user when a target activity is detected.