WO2014139416A1

WO2014139416A1 - Emergent abnormal event intelligent identification alarm device and system

Info

Publication number: WO2014139416A1
Application number: PCT/CN2014/073260
Authority: WO
Inventors: 黄鹏宇; 何跃凯; 周建雄; 彭元华; 郭振中
Original assignee: 成都百威讯科技有限责任公司
Priority date: 2013-03-11
Filing date: 2014-03-11
Publication date: 2014-09-18
Also published as: CN103839373A; CN103839373B; CN103198605A

Abstract

A intelligent identification alarm device for emergent abnormal events, comprises a video collecting module, a comprehensive information collecting module, an alarm sending module and a signal processing module. The video collecting module collects surrounding video information; the comprehensive information collecting module collects in real time surrounding signals, and transfers in real time the video information and the surrounding signals to the signal processing module. The invention is characterized in that the signal processing module processes and analyzes the images from the video information, and eliminates background images from said video information, so as enable identification of images from the video information. According to the result of the image identification, the signal processing module combines the surrounding signals and comprehensively analyzes same so as to identify whether the present event is an emergent abnormal event, and when distinguishing an emergent abnormal event is complete, thus decides whether to start the alarm sending module to send an alarm signal. Also provided is an intelligent identification alarm system for emergent abnormal events. The present emergent abnormal event intelligent identification alarm device and system are capable of comprehensive analysis and intelligent identification, and send alarm signals according to the identified emergent abnormal event.

Description

Instruction manual

Intelligent alarm device and alarm system for sudden abnormal events

The invention relates to an alarm system, in particular to an intelligent identification alarm device and system for sudden abnormal events. Background technique

We know that in order to prevent theft, anti-theft devices such as cameras are usually installed in residential areas or outdoor corridors, doors and windows, etc., for example, an intelligent video-based monitoring and alarm system for indoor personnel entering, mainly through the image of the camera. Intelligent analysis realizes the analysis of moving targets in indoor places, and the illegal intrusion of personnel reaches timely detection and alarm, while ignoring non-hazardous sports and reducing false alarms. There is also a home smart anti-theft system which is composed of at least a security door, a mechanical lock, a window, an open-air balcony and a balcony door. By installing on: (1) door net sensor on the security door, anti-theft door magnetic sensor, (2) lock tongue sensor mounted on the lock, key insertion/extraction sensor, single-bounce sensor, (3) mounted on the outside handle The handle position sensor in the assembly, at the same time, sets the balcony path detection area from the open balcony into the balcony door passage, and enters the window path detection area in the passage of the room from the window, thereby realizing automatic identification of the home and the thief in the home. system.

Existing door and window anti-intrusion systems have the following disadvantages:

1. Most products in the existing market have a single function and do not have comprehensive analysis capabilities.

2. A single anti-intrusion detection technology has its own defects. In some cases, it cannot work normally. Once it is not working properly, the entire detection system is in a paralyzed state, and the stability of the system is poor. For example: Intelligent video analysis and detection may be interfered by unrelated motions such as ambient light changes and specular reflections, resulting in a large number of false alarms; pyroelectric infrared detection is susceptible to interference from temperature, strong light, and complex environmental motion; vibration detection is susceptible to wind The external force causes the interference caused by the vibration of the door and window; the glass breakage detection is easily interfered by the noise of the external environment; the door magnetic detection requires the door and window to be strictly closed, and will not be used if the room needs ventilation. For example, in hot summer, the user often needs to open the window. ventilation.

3. The traditional alarm system, such as "home intelligent anti-theft system", involves a large number of various sensors. Installation and debugging are usually very complicated, and the alarm information is not intuitive, and the scene cannot be known.

4. The traditional alarm system can't work when the external power supply is cut off, and it can be used for illegal intruders.

5. It is impossible to carry out certain monitoring of legal persons in the room. For example, the elderly who often go out or live alone for their children need certain measures to monitor them to ensure that they can be rescued in time when abnormal situations occur, but there is no current This type of functionality.

Summary of the invention

The object of the present invention is to provide an intelligent identification device and system for sudden abnormal events, which solves the problem that the existing alarm system has a single function, a high false alarm rate, is easy to be interfered, cannot work normally, and cannot fall to a legal person such as an elderly person. The problem of violent attacks and sudden anomalies.

In order to solve the above technical problems, the present invention adopts the following technical solutions:

An intelligent identification and alarm device for sudden abnormal events, comprising a video acquisition module, an integrated information collection module, an alarm transmission module, and a signal processing module, wherein the video collection module collects surrounding video information, and the integrated information collection module collects surrounding signals in real time. And the video information and the surrounding signal real-time rumor processing module; wherein: the signal processing module processes and analyzes the image in the video information, and removes the background image in the video information, Realizing image recognition in the video information;

According to the image recognition result, the signal processing module combines the surrounding signals to comprehensively analyze and identify whether the current event is a sudden abnormal event, complete the resolution of the sudden abnormal event, thereby determining whether to activate the alarm sending module, and issue an alarm. signal.

The alarm device as described above is characterized in that:

A supplemental illumination module is further included. When the video acquisition module performs video acquisition, the supplemental illumination module is automatically activated when the ambient illumination is insufficient, and the auxiliary illumination source is provided for the video acquisition module.

An alarm device according to any of the preceding claims, characterized in that: a power management module is further provided, and the alarm device is powered, and the power management module is connected to an external power source and a battery;

When the external power supply is normally powered, the entire product uses an external power supply;

When the external power source is turned off, the battery is automatically enabled to supply power to the alarm device.

An alarm device according to any of the preceding claims, characterized in that:

The integrated information collection module includes one or more modules of an audio collection module, a heat release infrared detection module, a vibration detection module, a glass break detection module, a door magnetic detection module, a smoke detector module, and a gas detector module; The heat release infrared detecting module detects difference information of temperature and background temperature in the surrounding space, and converts the information into a voltage signal;

The vibration detecting module detects a vibration condition in a surrounding space and converts into a voltage signal;

The glass breakage detecting module and the door magnetic detecting module detect broken sound and door magnetic change information of the surrounding glass and convert into a voltage signal;

The ambient signal collected by the integrated information collection module includes voice information collected by the audio collection module, information detected by the smoke detector module and the gas detector module, the heat release infrared detection module, and vibration detection. The voltage signal detected by the module, the glass breakage detection module and the door magnetization detection module.

The alarm device as described above is characterized in that:

The light source of the fill light illumination module is annular;

The heat release infrared detecting module, the vibration detecting module, the video collecting module, the audio collecting module and the alarm sending module are arranged and arranged on an outer edge of the annular light source;

The signal processing module and the power management module are installed in the middle of the annular light source.

The glass breakage detection module and the door magnetization detection module respectively convert the detected information into a voltage signal to output to the signal processing module by wired or wireless communication.

The alarm sending module sends out an alarm signal sent by the signal processing module by using wired or wireless communication. The alarm device as described above is characterized in that:

The wireless communication method includes 3G, WIFK Zigbee. Bluetooth or other wireless communication methods.

The alarm device as described above is characterized in that:

The signal processing module collects signals and videos sent by one or more modules of the heat release infrared detection module, the vibration detection module, the glass break detection module, the door magnetic detection module, the smoke detector module, and the gas detector module. The signal transmitted by the acquisition module and the audio collection module is subjected to comprehensive analysis, thereby completing the resolution of the sudden abnormal event, and issuing an alarm signal through the alarm sending module.

The signal processing module includes a main processing chip that employs a dual core architecture mode in which the main core processor and the slave core processor are combined.

The alarm device as described above is characterized in that:

The main core processor receives the video information and the surrounding signals collected by the video collection module and the integrated information detection module;

Transmitting the video information and the surrounding signal to form a multi-source signal to the slave core processor;

The slave processor resolves the sudden abnormal event to form a final decision signal.

The alarm device as described above is characterized in that:

The distinguishing sudden abnormal event mainly includes running an audio and video intelligent analysis algorithm, and simultaneously distinguishing with other information in the surrounding signal.

The alarm linkage module is further included, and the alarm linkage module sends a warning to the sound and light alarm device in response to the alarm transmission module.

Also included is a voice output module that receives voice from the remote terminal and broadcasts the output. The invention also provides a sudden abnormal event intelligent identification alarm system, the system comprising a storage module and an alarm device as described above; the storage module establishes communication with the alarm device by means of wired or wireless communication Connecting, storing information collected by the alarm device; It is characterized by:

The distinguishing the abnormality event includes processing, analyzing, and identifying the image in the video information; and analyzing the surrounding signal that is synchronously collected by the integrated information collecting module;

The signal processing module completes the identification of the current event based on the processing, analysis, identification, and analysis of the surrounding signal in the video information, and realizes the resolution of the sudden abnormal event, thereby determining whether to activate the alarm. Send module, send out an alarm signal.

The alarm system as described above is characterized by:

The synchronization means that the video information and the surrounding signals are collected and analyzed simultaneously at the same time.

The alarm system as described above is characterized by:

The alarm system further includes a face database and a face registration module, and the feature of the standardized face image is extracted by the face registration module, and is entered and registered in the face database;

The processing, analysis, and identification of the image in the video information includes determining the identity of the person in the video information by face recognition.

The alarm system as described above is characterized by:

In the face recognition, the face recognition is performed by using the HAAR feature combined with the adaboost algorithm for face detection. The two-layer human eye locator is used to obtain the human eye positioning by the adaboost algorithm, and the face positioning result is used to correct the binoculars by image rotation. Horizontal face normalization;

Feature extraction is performed by two-dimensional Gabor filter, and the covariance distance between vectors is used as the matching metric. The nearest neighbor classification method is used to match the face image to be recognized with the face image in the face database.

An alarm system as claimed in any of the preceding claims, characterized in that:

The processing of the video information includes removing a background image in the video information, leaving a foreground image.

The alarm system as described above is characterized by:

Removing the background image in the video information includes the following steps:

A mixed Gaussian background model is used. Hybrid Gaussian background modeling assumes that the probability distribution of signal changes is fitted with K Gaussian distributions, expressed as

κ

Prob(x) = ω _ί η _ί (χ, μ _ί , σ _ι )

ί=1

Where ^ and σ are the mean and variance of the Gaussian distribution, and each Gaussian distribution "i ' ^ ) in the model is given a weight °^, where ^ and ^σ; respectively represent the mean and standard deviation of the Gaussian distribution;

The Gaussian distribution of multiple Gaussian distributions yields the probability distribution prob(x) of the signal, and the K Gaussian distributions are arranged in descending order, where the top Gaussian distribution

b

B = arg min _fc ω _η (χ) > Γ)

=1=1

Representing the background model, the remaining Gaussian distribution is the foreground model;

The pixel values of the video image information ^(and the front Β matching Gaussian distributions, if any, and wherein a Gaussian distribution of the pixel matching is successful as background pixels, the foreground image otherwise.

The identifying of the current event further includes removing an unrelated motion in the video information, tracking the target motion.

The alarm system as described above is characterized by:

The removing of extraneous motion in the video information includes employing a light interference filter to remove interference due to changes in light.

The alarm system as described above is characterized by:

The light interference filter removes interference by:

Through the background image modeling of the gray image, the motion region is ^Rg , and the region in the corresponding depth image is ^Rd . The real target region must satisfy the following conditions:

Wherein, and ^p _g are depth image and gray image motion detection results, respectively;

p ( ⁼¹ means the target motion pixel, ^p ( ^{= 0} means the background pixel).

The alarm system as described above is characterized by:

The alarm system also includes a face and head and shoulder detection filter to identify the action when the target is a person.

The alarm system as described above is characterized by:

The face and head and shoulder detection filters respectively use a haar wavelet combined with an adaboost classifier to implement face detection in a grayscale image, and a HOG feature combined with an SVM classifier to implement head and shoulder detection;

Recognizing as a target motion when the motion region detects a face feature or a head and shoulder feature;

Otherwise it is considered to be interference and filtered out.

After confirming the target action in the video information, the method further includes tracking the target action by the target association between the frames to form a motion track of the target, and providing a basis for the signal processing module to intelligently identify the event.

The alarm system as described above is characterized by:

The tracking includes:

Target position prediction, target feature selection, target association matching, and target feature update,

among them,

The target position prediction is to estimate the position of the target in the next frame according to the current frame and the position of the previous target, thereby improving the accuracy of the subsequent association matching, and using the following formula to complete the prediction of the position of the next frame.

Σ ) Σ (y _t -„ - y

V; =^ V =^

N_ and N _

Where ^ and ⁷ respectively represent the velocity of the target in the X direction and the y direction at time _t , N is the time window, and "and · ^ ^" respectively represent the abscissa and ordinate of the center of the target outer rectangle at time tn, the same as -" - ¹ denotes the abscissa and ordinate of the center of the outer rectangle of the target at t>nl, respectively;

The target feature selection is to select two features of the relatively stable and reliable target area S and the target circumscribed rectangular frame center C; the target association matching is to find the position of the previous frame target in the current frame, to achieve the target matching, and the target j at time t is

^0f , the target i is ⁰ " at time t-1, and the feature matching is performed when the following formula is satisfied:

Jsi(0 {C},0_ ₁ {C})<r where ^ is the search distance and needs to be determined according to the actual scene;

The matching criterion is as follows: arg min(« x (—— ^f ' ^f — ^u ) + (1-ω)χ < T r

Where ω is the characteristic weighting factor, taking the value of 0.5, 13⁄4 error upper limit, preventing mismatch, taking the value 0.4;

The target feature is updated by filtering out the mutation data, using the following formula

O _t {F}=axO _t _ ₁ {F}+(la)xO _t {F} F={S,Q

Among them, "for the update factor, take the value of 0.2 to achieve the update of the target feature.

28. An alarm system according to any of claims 15-27, characterized in that:

The alarm system can also realize the identification of a fall event, and the identification of the fall event includes the following steps: automatic three-dimensional modeling of the ground, setting the origin of the camera coordinate system on the image coordinate system, so that the translation matrix is reduced to

0

H

The conversion distance between the camera's optical center and the ground, the camera coordinate system and the world coordinate system are as follows

0

τ 0

H

R = R _x {a)*R _y (. )*R r ^ The angle between the three coordinate axes of the camera coordinate system and the three coordinate axes of the world coordinate system is ", and ^ assumes the origin of the image coordinate system and the camera coordinate system. If the origins coincide, then a point ^u , ^v , ^d in the depth image is

Ud vd , ,

(x = -, y = -, z = d)

The coordinates in the camera coordinate system are where ( ^w , ^v ) is the coordinates of the pixel in the image, ^d is the distance of the subject from the camera, and the focal length of the camera in the horizontal and vertical directions on the imaging plane; The point ^F ( ^x , y, ^z ) satisfies the following plane constraints in the camera coordinate system: ax + by + cz + d =0 where V = [a, b, (:: Γ is the normal vector of the ground plane, passing the normal vector It is possible to calculate the angle α of the three coordinate axes, β Y a _n bc

a = arccos - β = arccos - γ = arccos -

Ivi; , ivi

After obtaining a frame of depth image data, the plane fitting is performed by using the EANSAC algorithm, and multiple candidate ground planes D i, ^b i, ^c i, ^d i are obtained by fitting, and the plane is performed by the following prior knowledge. Coarse screening:

First, the ground occupies a large area in the image, that is, the real ground plane should include more image pixel points; 2. Normally, the angle Y between the camera and the z-axis is between 40 degrees and 80 degrees, and the angle a from the X-axis is between 0 degrees and 20 degrees, and the angle with the y-axis is substantially equal to 0 degrees.

In the plane remaining after the coarse screening, the plane farthest from the camera is selected as the ground plane, that is, it satisfies:

= arg max d _t

L≤i≤n

The distance from any point in the camera coordinates to the ground plane is:

Calculate the distance from the target center to the ground plane by the above formula;

The target is stationary, and the target is determined to be stationary by calculating the time domain difference of the grayscale image in the target area:

M = ^(x '^y>eR where & represents the gray value of the pixel point in the gray image at time t, and the target area of the knee is the pixel area of the target area;

When M < indicates that the target is stationary, where is a small positive number.

The alarm system can also realize the identification of a limb conflict event, and the identification of the limb conflict event is implemented by an optical flow vector and an audio analysis, and the optical flow vector analysis includes the following steps:

Optical flow vector analysis, the target region is obtained by target tracking, and the optical flow vector ^ ^, ^,... in the target region is calculated by using the optical flow of the locating point, and the region is realized by the amplitude weighted histogram ^H P ⁼ , ^h j , jD Statistical analysis of the optical flow vector, and then obtain the jth-order histogram ^hj by the following formula, h _j =C _h ∑A _v S(b(v _i )-j)

i=l The order value is 12, which is the normalization parameter, 'is the amplitude of the normalized optical flow vector ^V '', ^ ^V '') is the histogram interval corresponding to the optical flow vector ^V '', through the vector The direction is determined, ^(•) _Ko neckerdelta function, using the regional entropy ^E H to achieve a measure of severe irregular motion, the expression of ^E H is as follows:

It represents the j-th order amplitude weighted histogram. ^The larger the ^£ H is, the more severe the movement in the area is. The threshold T is set. When ^E H ^{> T} , the limb conflict is broken in the area.

Compared with the prior art, the beneficial effects of the present invention are as follows: The invention adopts intelligent audio and video analysis such as target behavior analysis, face recognition, voice recognition as the core technology, and combines advanced pyroelectric infrared detection, vibration detection, and audio. Internet of Things and wireless communication technologies such as detection, glass breakage detection, magnetic door detection and smoke detection, real-time dynamic detection and analysis of people in the indoor control area, long-term fall, illegal intrusion, theft, serious limb conflict, fire, gas leak, Sudden anomalous security incidents such as explosions, and the audio-visual or picture-based visual alarm signal is sent to the alarm receiving and processing terminal such as the household mobile phone or computer, the community monitoring computer, etc. through wired or wireless communication. The sound and light alarm is alarmed, and the voice output of the terminal device is received, and the voice intercom is realized. The invention also adopts a false alarm filter based on the sudden abnormal event review algorithm and the moving target feature analysis, and combines the multi-source detection information. , better solved the traditional indoor security system (such as Infrared detectors, intelligent anti-theft locks, etc.) have high false alarm rate and poor reliability. Finally, this product comes with a spare battery, which can supply itself for 2-24 hours without external power supply.

DRAWINGS

FIG. 1 is a schematic diagram of a connection of an embodiment of an indoor sudden abnormal event alarm system according to the present invention.

2 is a schematic diagram of hardware connection of an indoor sudden abnormal event alarm system according to the present invention.

FIG. 3 is a software structural diagram of a sudden abnormal event alarm system in the present invention.

4 is a schematic flow chart of a main processor end in a signal processing module of an indoor sudden abnormal event alarm system according to the present invention. FIG. 5 is a schematic flow chart of a slave processor in a signal processing module of an indoor sudden abnormal event alarm system according to the present invention. FIG. 6 is a schematic diagram of a face registration and face recognition process of the indoor emergency abnormal event alarm system of the present invention.

Fig. 7 is a schematic diagram showing the analysis of the face image of the face normalization in the indoor emergency abnormality alarm system of the present invention.

FIG. 8 is a schematic diagram of a fall detection process of the indoor emergency abnormal event alarm system of the present invention.

FIG. 9 is a schematic diagram of coordinate modeling for fall detection in the indoor sudden abnormal event alarm system of the present invention. FIG. 10 is a schematic diagram of a specific sound detection process of the indoor emergency abnormal event alarm system of the present invention.

FIG. 11 is a schematic diagram of a product experiment system connection of an indoor sudden abnormal event alarm system according to the present invention.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

1 shows an embodiment of an indoor sudden abnormal event alarm system of the present invention: an indoor sudden abnormal event alarm system, including

The video acquisition module uses a video sensor to collect a video stream signal, completes digitization of the image signal, and preprocesses the image signal to obtain a digital video signal that satisfies the requirements of the signal processing module and outputs the signal to the signal processing module;

The audio acquisition module completes the digital-to-analog conversion of the speech signal, the sampling and encoding of the speech signal, and the filtering process to obtain a digital audio signal required by the signal processing module and output to the signal processing module;

The heat release infrared detecting module senses the difference between the temperature of the moving object and the background object. When the human body moves, the pyroelectric infrared can sense the difference information between the human body temperature and the background temperature, and converts it into a voltage signal and outputs it to the signal processing module; the vibration detecting module Transforming the deformation or force information of the external vibration into a voltage signal and outputting it to the signal processing module;

The glass break detection module and the door magnetic detection module convert the corresponding information into a voltage signal and output to the signal processing module through wired or wireless communication, and the auxiliary signal processing module makes a corresponding judgment; The signal processing module collects the signals transmitted by the heat release infrared detection alarm module, the vibration detection module, the video acquisition module and the audio collection module, and performs comprehensive analysis to distinguish between an illegal invasion, a violent conflict, a long-term fall, a gas leak, and the like. An abnormal event, and an alarm signal is sent to the alarm sending module;

a wireless receiving module, configured to receive an alarm signal sent by the vibration detecting module, the glass breaking detecting module, the smoke detector, the emergency call button, and the alarm sensor on the door magnetic detecting module;

The alarm sending module adopts wired or wireless communication mode, and is mainly used for receiving an alarm signal sent by the signal processing module and sending an alarm message to the mobile phone or the cell management center of the household;

Alarm linkage module, linkage sound and light alarm equipment work at the same time;

a voice output module that implements voice output from a remote terminal device such as a mobile phone or a monitoring center;

The infrared LED fill light module provides an auxiliary light source to the video capture module when the ambient illumination is insufficient, so that it can still collect valid image data in a low illumination environment;

The power management module is connected to the external power supply and the battery. When the external power supply is normally powered, the whole product uses an external power supply. When the external power supply is cut off, the internal backup battery is automatically enabled to ensure that the above modules continue to work normally after the external power supply is interrupted.

1 also shows a preferred embodiment of the indoor sudden abnormal event alarm system of the present invention, the system is installed in a casing, the video acquisition module, the audio collection module, the heat release infrared detection alarm module, and the vibration detection module. The alarm sending module, the wireless receiving module, the alarm linkage module and the alarm sending module are installed in a casing, and the signal processing module and the power management module and the switch light control module are installed in the middle of the casing, and the infrared LED fill light module is set. Between the outer edge of the casing and the middle.

2 shows another embodiment of the indoor emergency abnormality alarm system of the present invention. The core processing chip in the signal processing module adopts a dual-core architecture mode, that is, a "main processor + slave processor" dual-core architecture mode, and the main processor is mainly Complete audio and video signal collection, audio and video coding, pyroelectric infrared alarm signal collection, receiving vibration detection module, glass break detection module, door magnetic detection module, smoke detector, emergency call button and other distributed alarm sensors through wireless receiving module The sent alarm signal transmits the multi-source alarm signal together with the audio and video data to the slave processor; the slave processor is mainly used to run the audio and video intelligent analysis algorithm, and judges whether there is a sudden abnormal event through video and sound, and simultaneously combines pyroelectricity Infrared alarm signal, vibration alarm signal, glass break alarm signal, door magnetic alarm signal and other multi-source information form the final decision signal. If it is determined that a sudden abnormal event occurs, send an alarm signal to the main processor and activate the sound and light alarm. Alarm.

As shown in FIG. 3, the software of the indoor emergency abnormal event alarm system of the present invention is mainly composed of a main processing process and an intelligent processing process, and the main processing process is composed of a signal acquisition module, a voice output module and an alarm sending module. Running on the ARM side; intelligent processing process, consisting of intelligent analysis module and multi-source information fusion false alarm filtering module, running on the DSP side. The signal acquisition includes audio and video acquisition module, pyroelectric infrared alarm signal acquisition module and wireless alarm signal acquisition module, which is responsible for the acquisition of audio and video signals and voice output. In order to ensure that the collected signals can be synchronized for subsequent processing, appropriate delay processing is performed on various alarm signals to synchronize with the audio and video acquisition signals. The intelligent processing process completes intelligent audio and video analysis, multi-source information fusion and false alarm filtering, and sends an alarm signal to the main processing process. The main processing process sends an alarm message to the external terminal or activates the linkage alarm device alarm through the alarm sending module.

As shown in Figure 4, the main processing software flow: After the system is powered on, the initialization of the main processor program is completed first, then the peripheral acquisition device is completed, the communication device is initialized, and the slave processor program is started. Create an audio and video capture thread, a pyroelectric infrared detection thread, a wireless receiving thread, a slave processor communication thread, and an alarm sending thread.

Audio and video collection thread: When receiving audio and video data, combine one frame of multi-source data, including video data, audio data and other various sensor alarm identifiers, put them into Rame Buffe r for communication thread to call, and complete audio and video pre-processing. Recorded to provide alarm information.

Pyroelectric infrared detection thread and wireless receiving thread: Receive alarm signals Ai of various alarm sensors, wherein Ai can be pyroelectric infrared alarm signal, vibration alarm signal, glass broken alarm signal, door magnetic alarm signal and so on. When the alarm signal is received, the corresponding alarm flag H is set to 1, indicating that the i-th sensor has an alarm, and the counter Ci is set to an initial value. Value, otherwise the counter Cii is decremented. When the counter is reset to zero, the alarm flag H is set to 0. The delay counter is set here to ensure that the alarm information of various sensors can be synchronized.

From the processor side communication thread: Receive an alarm signal sent from the processor side to notify the alarm sending thread. If there is one frame of multi-source data in the Rame Buffer, the data is sent to the slave processor.

Voice output / alarm sending thread: After receiving the alarm signal, organize related alarm information, such as audio and video of the quotation, pictures, etc., send to the external terminal, and start the equipment such as sound and light alarm. If the voice output information is received, the voice output function is immediately turned on, and the remote terminal is connected to complete the voice output.

As shown in Figure 5, the intelligent processing software flow: First, after the main processor side main processing process is started, the multi-tasking kernel and the intelligent processing process are added to the slave processor memory, and the slave processor completes a series of initializations to automatically create a communication. Threads and analysis threads. The communication thread first determines whether it is necessary to send an alarm signal, and if so, sends an alarm signal to the main processing process, otherwise it determines whether a multi-source information is received, wherein the multi-source information includes: audio and video data and alarm information of other sensors, which will be more The source information is placed in the frame buffer RameBuffe r, and a receiving confirmation signal is sent to the main processing process, indicating that the intelligent processing process works normally. The analysis thread reads a frame of data from Rame Buffe r, extracts audio and video data from it for analysis, and obtains the analysis result, and combines other alarm information in the data frame to complete multi-source data fusion and false alarm filtering to determine whether an alarm needs to be sent. signal.

FIG. 6 is another embodiment of the indoor emergency abnormality alarm system of the present invention. The household identity determination includes the identification of the household owner by face recognition, wherein the face recognition is mainly divided into:

Face registration, collecting positive face photos of different angles of the same person, finding the position of the face in the image through face positioning, and normalizing the face size, face angle and illumination, extracting the features of the standardized face image, and entering Register a user database;

Face recognition, using HAAE sign combined with a dab oo st algorithm to realize face detection to achieve face localization, using a two-layer human eye locator, all obtained eye positioning by a dab 0 0 st algorithm, using face positioning results The face is normalized by image rotation to correct the face, the feature extraction is performed by the two-dimensional Gab 0 ι3⁄4 chopper, and the covariance distance between the vectors is used as the matching metric. The face image and database of the query are realized by the nearest neighbor classification method. Matching of face images. The Adab o o st algorithm combines a large number of weak classifiers with general classification ability into a strong classifier according to the decline of the training error index. The HAAE sign provides a large number of weak classification features for the adab o o st algorithm, which ensures that the adab o o st algorithm generally finds weak classifications with excellent performance. In the implementation process of face detection, the use of integral histogram and cascade classifier greatly reduces the processing time while ensuring higher detection accuracy; human eye positioning is generally divided into two layers, wherein the first layer is coarse positioning. The positioning area selects most of the eye area including the eyebrows, the second layer is the precise positioning, and the positioning area only contains the eye area. Compared with the precision positioner, the coarse positioner has more regional information, so the stability of the positioning is higher, there is basically no large positional deviation, and the precise positioner can achieve precise positioning of the human eye, but is susceptible to The interference of the eyebrows and the corners of the eyes caused a positioning error. The approximate position range of the human eye is determined by the geometric proportional relationship on the coarsely positioned ^ ftH, within which the precise positioning of the human eye is achieved using a precise positioner. Through the coarse to fine positioning method, the influence of eyebrows and eye corners on positioning is reduced, and the accuracy of positioning is improved. Face standardization is a very important step in face recognition, and the standardization result directly affects the quality. The accuracy of face recognition. The face standardization mainly completes the geometric correction and brightness correction of the face image. It is easy to realize the geometric correction of the face image by using the result of the human eye positioning in the previous step. First, the eyes are corrected to the horizontal by the image rotation, and the face image is intercepted by the distance d between the eyes. As shown in Figure 7, the image is finally scaled to 80x80 pixels.

The brightness correction mainly eliminates the influence of uneven illumination on subsequent recognition to some extent. It mainly includes the illumination surface fitting, deducting the illumination surface, histogram equalization and gray value normalization to zero mean, unit variance. It is assumed here that the illumination surface is a plane. The point on the light surface satisfies the following formula: ^{/5 =} + ^ ^{+ (} written in matrix form ie " ^{3⁄4 =} , where column vector representing the gray value of the pixel of the image is arranged, N represents the coordinates corresponding to the pixel point, One column represents the abscissa, The second column represents the ordinate, the third column is filled with 1, ^{p =} [ ^{ab C} F. The plane parameters a, b, c can be obtained by linear regression, _? Ρ =, Ν ^Τ ΝΠ

The Gab 0 r wavelet feature can be selected, and the Gab o r transform has excellent performance in analyzing the local texture of the image. Two-dimensional Gab on filter ^ (can be expressed as:

Where is the image coordinates, which is the center frequency of the filter, and y are the projections on the horizontal and vertical axes, respectively.

Shadow, % is the direction of the filter, "and ^v represents different values,

For Gaussian envelope, ^ex P( ^fc ) is a complex-valued plane wave. The two-dimensional Gab 0 waver modulates the sine wave plane of a specific frequency and direction by a two-dimensional Gaussian function, and realizes the analysis of image textures of different scales and different directions by changing the frequency and direction of the sine wave plane.

A face image of 80x80 size is obtained by face normalization. Here, five filter scales and eight filter directions are selected to obtain 40 Gab 0 waves of different directions and frequencies, and a face image is passed through the filter roll. After the product, 40 amplitude images of Gab 0 r wavelet transform are obtained, and the obtained Gab o 1# dimension is 163840. Such a high-dimensional feature vector greatly reduces the speed of identifying the classification, so it is necessary to reduce the dimension of the feature vector. Here, 4x4 uniform downsampling is used to achieve feature dimensionality reduction.

The nearest neighbor classification method is used to match the face image of the query with the database face image, and the covariance distance between the vectors is used as the matching metric. At the same time, the reliability of the final matching can be measured by the covariance distance.

According to another embodiment of the indoor emergency abnormality alarm system of the present invention, the signal processing module is connected with a storage module (Storage), and different types of storage media are used for different application situations, and SD and TF cards are controlled by SDIO. This type of storage medium is easy to replace; Nand is controlled by Nandfksh, this type of integration is high, but it is not easy to replace the storage medium; SSD is controlled by PCI-E or S VIA interface, this type of storage space can be very large;

Set up with Ethernet interface (RJ45);

An external power supply (DC IN) and a battery (Batter) are connected through a power management module (Powermanage r); and a memory (DDR) for controlling the operation of the CPU is connected;

Connected to a flash (Hash) for storing system boot programs, configuration parameters, and log information;

Connected with wireless module (3G, W1FX) for audio and video data and remote control signal transmission; connected with wireless module (zigb ee, Hue to oh or other wireless module), used to receive 4 分布式 sent by distributed alarm Alarm signals, such as glass break signals and door open signals, can also send control information to other devices.

The linkage alarm module is connected with an external alarm light, a siren linkage device, a local sound and light alarm, or other related equipment.

8 and 9 illustrate a preferred embodiment of the indoor sudden abnormal event alarm system of the present invention. The video capture module includes an ordinary camera detection module and a depth camera detection module in which image data are aligned with each other, for a point P in the image. The six-dimensional information component ^, ^^, ^", ^^> is obtained, where <"^^> is the coordinate of the pixel point in the image, d is the distance of the subject from the camera, and <^^> is the color component. The method for detecting the moving target and detecting the sudden abnormal event alarm system in this room is as follows:

The color image is converted into a gray image, and the background model is established for the gray image and the depth image respectively. The background model is established by using mixed Gaussian. The mixed Gaussian background model assumes that the probability distribution of the signal change can be fitted by a Gaussian distribution, expressed as

κ

ρΓθΗχ) = ^ ω _ί η _ί (χ, μ _ί , σ _ί ) where; "and σ is the mean and variance of the Gaussian distribution, each Gaussian distribution C^'A' in the model) is given a weight, where ^A and ^σ '' respectively represents the mean and standard deviation of the Gaussian distribution, and the Gaussian distribution obtains the probability distribution of the signal by linear combination. One Gaussian distribution is arranged in descending order of *^°", and the Gaussian distribution of the front ranks enough to represent the distribution of the background. At the same time, the mixed Gaussian model can automatically maintain the changes of the scene, and at the same time, the error can be corrected by learning to correct the error.

The first Gaussian distribution is used as the background model, and the remaining Gaussian distribution is considered to be the foreground.

B = arg min _fc ω _η (χ) > Τ) Match the current pixel value ^ with the previous Gaussian distribution. If any of the Gaussian distributions match successfully, the pixel is the background pixel, otherwise it is the foreground. Motion pixel, matching method is as follows

\ Ι(χ) - μ _ί (x) l< c χ c _; (x) i = l .· ·· · B .

The target confirms that the moving object is detected by the background modeling method, but many movements are caused by the change of indoor light, the swing of the curtain, the flashing of the TV picture, etc., and the monitoring object here is a person, so the person needs to be The movement is separated from a large number of unrelated movements. Here, through the light interference filter, the face detection filter and the head and shoulder detection filter, the gray image analysis is mainly used, and the depth image analysis is supplemented, and the face detection and the head and shoulder detection are combined to realize the monitoring target confirmation.

The light interference filter: the change of the light causes the gray value of the image to change. When the speed and amplitude of the gray value change exceeds the adaptation range of the background model, it is detected as a moving target, and the light image is filtered by the depth image. Motion, depth image is based on the infrared light detection method, which is basically free from the interference of illumination changes. The motion region is ^Rg through the background image modeling of gray image, and the region in the corresponding depth image is ^Rd . The real target region must satisfy the following conditions. condition:

^ >τ

Where (and ^p _g (respectively for depth image and grayscale image motion detection results, ^p ) = ¹ for motion image Prime, ^P ^) ^{= ()} represents the background pixel;

Face and head and shoulder detection filter: There are a lot of real movements in the actual monitoring scene, such as the rotation of the fan, the swing of the curtain, the flickering of the TV picture, the movement of the pet, and the present invention only cares about the movement of the person, the person is different from The obvious appearance features of these movements, such as face features and head and shoulder features, here in the gray image using haar wavelet combined with the adab oo st classifier for face detection, HOG features combined with SVM classifier to achieve head and shoulder detection, when in motion When the area detects a face feature or a head and shoulder feature, it is described as a person who is exercising, otherwise it is considered to be interference and filtered out;

Target tracking, based on the target confirmation, forms the target's motion trajectory through the target association between frames, and provides a basis for subsequent target behavior analysis. Target tracking mainly includes: target location prediction, target feature selection, target association matching, and target feature update, among which

The target position prediction is to estimate the position of the target in the next frame according to the current frame and the position of the previous target, which is beneficial to improve the accuracy of the subsequent association matching. Here, the method of estimating the target speed is used, and the position is predicted by the following formula.

Σ ( ^Χ , -η - ^X tnl ) Σ (y _t -n - -„-1 )

Where ^V f and ⁷ respectively represent the velocity of the target in the x direction and the y direction at time t, N is the time window, and "and · ^ ^" respectively represent the abscissa and ordinate of the center of the target outer rectangle at time tn, and the same - "- ¹ denotes the abscissa and ordinate of the center of the outer rectangle of the target at t>nl, respectively;

The target feature selection is to select two features of the relatively stable and reliable target area S and the center C of the target circumscribed rectangular frame. The target association matching is to find the position of the previous frame target in the current frame and achieve the target positioning. According to the correlation tracking principle, as long as sufficient image sampling rate is ensured, the position change of the same target between adjacent frames will not be Too large, so the target search range can be limited to a small distance range, and the risk of mismatching is greatly reduced. The target j is ^Qf at time t, and the target i is ⁰ " at time t-1. Feature matching only when:

J si(0 {C}, 0 _ ₁ {C}) < r where ^ is the search distance and needs to be determined according to the actual scene.

The matching criterion is as follows: arg min(« χ (—— ^f ' ^f — ' ) + (1 - ω) χ ~ '- ' ) < T where w is the weighting factor of the feature, which is 0.5, Ί3⁄4 upper error limit, Prevent mismatch, value 0.4

The update of the target features is performed by filtering out the mutated data to ensure that the data does not appear to have large fluctuations in the first-order smoothing manner. For the specific implementation, see the formula for the next line.

q {F} =ax O _tl {F} +(la)xO _t {F} F = {S, Q

Where "is the update factor, value 0.2

The indoor sudden abnormal event alarm system is used to detect the fall as follows, the ground automatic three-dimensional modeling, will be with the video The camera connected to the acquisition module is tilted down at an angle to the ground, and the origin of the camera coordinate system is set to the image.

0

H

On the Z axis of the coordinate system, the translation matrix T is simplified to , where H represents the distance from the camera's optical center to the ground, and the conversion formula of the camera coordinate system and the world coordinate system is as follows

0

T 0

HR = R _x (a)*R _y (fi)*R

The angle between the three coordinate axes of the camera coordinate system and the three coordinate axes of the world coordinate system ^a J . Assuming that the origin of the image coordinate system coincides with the origin of the camera coordinate system, a point (", ^v , ) in the depth image is in the camera coordinate system. Lower coordinates

Ud vd , ,

Where ( ^w , ^v ) is the coordinates of the pixel in the image, d is the distance of the subject from the camera, and is the focal length of the camera in the horizontal and vertical directions on the imaging plane; set the point y, ^z ) on the ground The following plane constraints are satisfied in the camera coordinate system: ax + by + cz + d =0 where = ^[Ω , ^c f is the normal vector of the ground plane, and the angle ^{α of the} three coordinate axes can be calculated by the normal vector, and Ϊ■ a _n bc

a = arccos - β = arccos - γ = arccos -

Ivi; , ivi

After obtaining a frame of depth image data, the EANSAC algorithm is used to achieve plane fitting. By fitting, multiple candidate ground planes ^^ , ^', ⁶ ^ ⁶ ^ can be obtained, and the plane is coarsely filtered by the following prior knowledge. ;: First, the ground occupies a large area in the image, that is, the real ground plane should include more image pixel points; Second, the angle between the camera and the z-axis is usually 40 degrees and 80 degrees. The angle ^a between the axis and the X axis is between 0 and 20 degrees, and the angle with the y axis is substantially equal to 0 degrees.

= arg max d _i

1

The distance from any point in the camera coordinates to the ground plane is:

, I ax

La ² +b + c Calculate the distance from the target center to the ground plane by the above formula;

The target is judged statically, and the target is still determined by calculating the time domain difference of the grayscale image in the target area: ) g _t x, y) - g _t _ _l x, y) \

M = ^(x '^y>eR where & ^(JC , which represents the gray value of the pixel point in the gray image at time t, and R represents the target area, which is the pixel area of the target area. When M < indicates that the target is stationary, Where ε is a small positive number.

In general, firstly, the ground 3D modeling is performed to determine the ground area in the image, and the 3D coordinates of the ground are obtained. On the basis of the target tracking, the distance between the center of the target and the ground plane is calculated, if Η is smaller than the set The value of 阔 indicates that the person’s body is close to the ground. If the time when the target is detected to be stationary is greater than a certain threshold, the fall report is triggered.

FIG. 10 shows another preferred embodiment of the indoor sudden abnormal event alarm system of the present invention. The audio collection module also detects a specific sound, and the detection method package is as follows: Voice signal preprocessing, assuming the sampling rate of the audio signal X ( For the value 83⁄4Ήζ, X (subsequently pre-emphasis, framing and windowing, the window function selects the Hanning window, and removes the mean, avoiding the DC component affecting the line near w = 0;

Feature extraction, using the periodic graph method in classical spectral estimation, using IFT, finally obtaining a normalized power spectrum

^{Χ η} , , Extract a number of 24 Mel filter banks. The power spectrum is filtered by the Me l filter bank to obtain the logarithm, and then the discrete cosine transform is used to obtain the Mel cepstrum MFC C coefficient. The Mel filter bank is composed of a set of triangular bandpass filters distributed according to the Mel frequency standard;

The training of GMM audio identification model adopts the maximum expectation: EM algorithm to obtain the GMM model, given the training sample set ^Χ = { ^χ ι, ^χ 2,···, ^χ _η } , the likelihood function of GMM is

Where model parameters =

Pi represents the probability of the Gaussian model, ^w ''' ^∑ '' represents the mean vector and covariance matrix of the Gaussian model, respectively.

The EVI algorithm consists of two steps, E step to obtain the expectation, calculate the auxiliary function β( , ), maximize the expectation of the step, and maximize the β(Λ^).

y

Where X is the observed value and Y is the implicit state. When the expected maximum values calculated by two adjacent iterations are not much different, the algorithm converges and stops iterating, as shown in the following equation:

Q _t ( , )-Q _t ( ,i)<s

Where t represents the number of iterations, which is a smaller positive number;

Using the training model to identify, the model parameters are obtained through GMM training. The speech segments are extracted and sent to the GMM model to calculate the similarity. The similarity is used to judge the category of the speech segment.

According to an embodiment of the indoor sudden abnormal event alarm system of the present invention, a method for detecting a limb conflict in an indoor sudden abnormal event alarm system includes optical flow vector analysis and audio analysis, accompanied by severe irregularity in an outbreak of limb conflict The movement and loud whistling can detect limb conflicts by optical flow vector analysis and audio analysis. When both methods detect limb conflicts, they trigger a limb conflict alarm and capture a live photo.

Optical flow vector analysis, the target area can be obtained by target tracking, and the optical flow vector ^V = ^{ν ι, ^ν2 , ···, in the target area is calculated by using the MT feature point optical flow, and the amplitude weighted histogram is used ^{= {h} j = i, 2, .., n realizes the statistical analysis of the regional optical flow vector, and then obtains the jth order ^kj , h^C^Sibiv,)-])

Here, the order can take the value 12, ^^ is the normalization parameter, 'is the amplitude of the normalized optical flow vector ^V '', and ( ^ν '') is the histogram interval corresponding to the optical flow vector ⁷ ', through the vector The direction is determined, ^(·) is the Koneckerdelta function, and the regional entropy is used to achieve the measure of severe irregular motion. The expression is as follows:

Where ^ denotes the jth order amplitude weighted histogram. A larger one indicates that the movement in the area is more severe and irregular, and the threshold T is set, indicating that a physical conflict has erupted in the area.

The entire workflow of the present invention:

The user can turn the alarm function on or off as needed. The detection method can select one or several of video detection, audio detection, pyroelectric infrared detection, vibration detection, glass breakage detection, smoke detection, and magnetic detection. Falling and serious attitude conflicts are detected by audio and video. Pyroelectric infrared detection, vibration detection and glass breakage detection are used in the window area. Video detection, audio detection, pyroelectric infrared detection, vibration detection and door magnetic detection are used in the area where the door is located. The alarm information can be one or more of video, audio or picture.

(1) After the device is powered on or reset, the signal processing module adds the operating system and application program from HASH, completes the initialization of the main processing chip and the configuration of the peripheral hardware, and then completes the initialization of each subsystem, and finally enters the normal state. Working status. When you use it for the first time, register your family member's face.

(2) The main processor end of the main processing chip continuously collects the audio and video signals and the pyroelectric infrared detection signals of the monitoring site, and simultaneously receives the alarm signals of other distributed alarm sensors through the wireless receiving module, and sends the multi-source data into the processing. The analyzer performs analysis and simultaneously performs audio and video pre-recording. If an alarm or warning signal is received from the processor, the pre-recorded audio and video, together with the captured photo, is sent to the cell monitoring center or the mobile phone of the household owner through the alarm signal sending module, and the alarm device such as the sound and light alarm is activated. .

(3) The main processing chip runs the intelligent audio and video analysis algorithm from the processor side, and analyzes from the perspective of video and audio respectively to find out whether there are sudden abnormal events such as falling for a long time, serious limb conflict, strangers entering the room illegally. Occurs, combined with other sensor alarm information, and uses decision fusion technology to obtain the final decision result, and sends the result to the main processor.

(4) After receiving the alarm, the head of the household or the monitoring center can confirm the alarm information sent back, or open the voice intercom function, and talk with the indoor personnel to further understand the occurrence of sudden abnormal events.

Product experiment

Experimental environment and equipment

Experimental site: 120 square meters 2 bedrooms 2 halls residential homes.

Experimental equipment: (1) 1 camera carrying a sudden abnormal event alarm system; (2) 1 wireless magnetic detector; (3) 2 wireless glass break detectors; (4) Wireless smoke detector 1 (5) One wireless gas leakage sensor; (6) One mobile phone that installs the emergency abnormality alarm receiving processing software.

As shown in Figure 11, a ceiling-type emergency alarm camera is installed in the living room, and a glass break detector, a gas leak detector, a smoke sensor, a door magnetic detector, and a wireless detector are installed in the bedroom, the kitchen, and the entrance door respectively. Communicate with the alarm camera, and connect the alarm alarm through the output port of the alarm camera 10.

Test methods and results: sudden abnormalities test face method

test results

Condition Condition

Simulated old people in the living room during the day, the siren screamed 101 times. Detection accuracy: between the room, the corner of the house, the door and window, etc. The phone receives the sound, Figure 98%;

The position fell 100 times, view the hand piece alarm: 101 times, its false positive rate: 3%; the machine receives the alarm situation. Misreported 3 times, underreported 2 Missing rate:

Times. 2% simulated old man night (light weak) Siren tweet 104 times. Detection accuracy: Turn on audio and video, drop

In the middle of the living room, in the corner of the house, the phone receives the sound, Figure 97%;

Inverted, infrared detection,

Falling down by doors and windows, etc. 100 pieces of alarm: 104 times, its false positive rate: 4%;

Interference filtering algorithm.

Times, check the mobile phone to receive the alarm. 4 false positives, false negatives 3 false negative rate: situation. Times. 3% Turn on the indoor TV, make the window, and the siren will sound 55 times. Detection accuracy: The curtain flutters, the dog runs indoors, the phone receives the sound, the picture is 94%;

Motion and other interference phenomena. At the same time, the module alarm: 55 times, of which the false positive rate: 95%; the old man falls: 50 times. False positive 5 times, underreported 3 times false negative rate:

6% severe physical rushing to start severe physical examination Simulated personnel conflict: 40 times The siren screamed 41 times. Detection accuracy: Detecting algorithm, vibration detection, receiving sound, mobile phone, figure 100%;

Measurement, infrared detection, film alarm: 41 times, of which the false positive rate: 2.5%; interference filtering and other miscalculations 1 time, missing 0 times false negative rate: law. 0% Illegal intrusion turns on intrusion detection, and the simulated thief screams 52 times from the door and window. Detection accuracy: Vibration detection, infrared into the room: 50 times. The phone receives the sound, the picture is 100%;

Detection, interference filtering, slice alarm: 52 times, of which the false positive rate: 4%; and other algorithms. False positive 2 times, false negative 0 times false negative rate:

0% Smoke Detection Turn on the smoke detector Simulate smoke indoors: The siren screams 19 times. Detection accuracy:

20 times. The mobile phone receives the smoke SMS report 100%;

Police: 19 times, of which false positives false positive rate: 0%; 0 times, missed one time, false negative rate:

5% Gas Leakage Open Gas Leakage Simulation Simulate indoor gas production. The siren screams 10 times. Detection accuracy: Tester Leak: 10 times. The mobile phone receives a gas leak of 100% short;

Letter alarm: 10 times, of which false positive rate: 0%; false positive 0 times, missing 0 times false negative rate:

0% glass breakage open glass breakage simulation window glass broken: 10 sirens 10 times. Detection accuracy: Tester times. Mobile phone connected glass broken text message 100%;

Alarm: 10 times, of which false positive rate: 0%; reported 0 times, missed 0 times, false negative rate:

0% door magnetic detection Open door magnetic detection Simulate open the door to the home: 100 sirens 98 times. Detection accuracy: times. The mobile phone access door opens 98%;

SMS alarm: 98 times, its false positive rate: 0%; misreported 0 times, missed 2 times, false negative rate:

2% Explosion Turn on audio and video detection. Simulate indoor household appliances explosion. The siren screams 20 times. Detection accuracy: Measurement, vibration detection Sound: 20 times. Mobile phone connected to the room explosion SMS 100%;

law. Alarm: 20 times, of which the false positive rate is: 0%;

0 times, missed 0 times, false negative rate:

0% emergency button Turn on emergency call function Press the emergency button at intervals: 50 Emergency call: 20 Detected accurately: times. Times, of which 0 is falsely reported and 100% is missed;

Report 0 times false positive rate:

0%;

Reporting rate: 0% Test conclusion: In the indoor environment, the test results are consistent with the test expectations, and the function and performance are in full compliance with the product design requirements.

The invention has the following advantages:

Advantage 1: Strong function and wide application. The product can detect almost all abnormal events in the room, especially the high-precision fall detection can be widely used for the supervision of elderly people living alone.

Advantage 2: The false positive rate is low. Built-in powerful false alarm filtering algorithm, which filters the sudden abnormal events caused by indoor ambient light changes, curtains, and pets.

Advantage 3: The alarm is timely, the alarm information is intuitive and flexible, and it can be audio, video or picture.

Advantage 4: The equipment is easy to use, install and maintain. With wireless transmission, the wiring is simple and easy to use and maintain. Advantage 5: It can be used without interruption. Even if the power is cut off, the built-in battery can be automatically turned on and continue to be used.

Although the present invention has been described herein with reference to the various embodiments of the present invention, it is understood that many modifications and embodiments may be Within the scope and spirit of the principle. More specifically, various variations and modifications can be made in the component parts and/or arrangements of the subject combination arrangement within the scope of the present disclosure, the drawings and the claims. In addition to variations and modifications to the components and/or arrangements, other uses will be apparent to those skilled in the art.

Claims

Rights request

1. An intelligent identification and alarm device for sudden abnormal events, including a video collection module, a comprehensive information collection module, an alarm sending module, and a signal processing module. The video collection module collects surrounding video information, and the comprehensive information collection module collects surrounding video information in real time. signal, and transmit the video information and the surrounding signals to the signal processing module in real time; it is characterized in that: the signal processing module processes and analyzes the images in the video information, and converts the background image in the video information into Remove to achieve image recognition in the video information;

According to the image recognition result, the signal processing module combines the surrounding signals to comprehensively analyze and identify whether the current event is a sudden abnormal event, complete the resolution of the sudden abnormal event, and thereby decide whether to activate the alarm sending module and issue an alarm. Signal.

2. The alarm device according to claim 1, characterized in that:

It also includes a supplementary lighting module. When the video collection module performs video collection, the supplementary lighting module automatically starts when the ambient illumination is insufficient to provide an auxiliary lighting source for the video collection module.

3. The alarm device according to claim 1 or 2, characterized in that:

It also includes a power management module to provide power to the alarm device, and the power management module is connected to an external power supply and a battery;

When the external power supply supplies power normally, the entire product uses the external power supply;

When the external power supply is cut off, the battery is automatically activated to provide power to the alarm device.

4. The alarm device according to any one of claims 1-3, characterized in that:

The comprehensive information collection module includes one or more modules among an audio collection module, a thermal release infrared detection module, a vibration detection module, a glass breakage detection module, a door magnetic detection module, a smoke detector module, and a gas detector module; wherein, The heat release infrared detection module detects the difference information between the temperature in the surrounding space and the background temperature, and converts it into a voltage signal;

The vibration detection module detects vibration in the surrounding space and converts it into a voltage signal;

The glass breakage detection module and the door sensor detection module detect the breaking sound of the surrounding glass and the door sensor change information and convert them into voltage signals;

The surrounding signals collected by the comprehensive information collection module include voice information collected by the audio collection module, information detected by the smoke detector module and gas detector module, the heat release infrared detection module, vibration detection The voltage signal detected by the module, glass breakage detection module and door magnetic detection module.

5. The alarm device according to any one of claims 1 to 4, characterized in that:

The signal processing module includes a main processing chip, and the main processing chip adopts a dual-core architecture mode that combines a main core processor and a slave core processor.

6. An intelligent recognition and alarm system for sudden abnormal events, the system includes a storage module and an alarm device as claimed in any one of claims 1 to 5; the storage module communicates with the alarm device through wired or wireless communication Establish a communication connection and store the information collected by the alarm device;

Its characteristics are:

The distinguishing of sudden abnormal events includes processing, analyzing, and identifying images in the video information; analyzing the surrounding signals synchronously collected by the comprehensive information collection module;

The signal processing module completes the identification of current events based on the processing, analysis, and identification of images in the video information and the analysis of the surrounding signals, and realizes the resolution of sudden abnormal events, thereby deciding whether to activate the alarm The sending module sends out an alarm signal.

7. The alarm system according to claim 6, characterized in that:

The synchronization means that the video information and the surrounding signals are collected at the same time and analyzed synchronously.

8. The alarm system according to claim 6, characterized in that:

The alarm system also includes a face database and a face registration module. The features of the standardized face image are extracted through the face registration module and entered and registered in the face database;

The processing, analysis, and recognition of images in the video information include determining the identity of the person in the video information through face recognition.

9. The alarm system according to claim 8, characterized in that:

In the face recognition, HAAR features are used combined with the adaboost algorithm for face detection to achieve face positioning; a two-layer human eye locator is used to obtain the human eye positioning through the adaboost algorithm, and the face positioning results are used to pass the image Rotation corrects the eyes to be horizontal to normalize the face;

Feature extraction is performed through a two-dimensional Gabor filter, the covariance distance between vectors is used as a matching measure, and the face image to be identified is matched with the face images in the face database through the nearest neighbor classification method.

10. The alarm system according to any one of claims 6-9, characterized in that:

The processing of the video information includes removing the background image in the video information, leaving a foreground image.