CN116443682B - Intelligent elevator control system - Google Patents

Intelligent elevator control system Download PDF

Info

Publication number
CN116443682B
CN116443682B CN202310728163.0A CN202310728163A CN116443682B CN 116443682 B CN116443682 B CN 116443682B CN 202310728163 A CN202310728163 A CN 202310728163A CN 116443682 B CN116443682 B CN 116443682B
Authority
CN
China
Prior art keywords
model
training
elevator
module
representing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310728163.0A
Other languages
Chinese (zh)
Other versions
CN116443682A (en
Inventor
胡景博
姚文磊
马宁
王炜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Lanzhou University of Technology
Original Assignee
Lanzhou University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Lanzhou University of Technology filed Critical Lanzhou University of Technology
Priority to CN202310728163.0A priority Critical patent/CN116443682B/en
Publication of CN116443682A publication Critical patent/CN116443682A/en
Application granted granted Critical
Publication of CN116443682B publication Critical patent/CN116443682B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B1/00—Control systems of elevators in general
    • B66B1/24—Control systems with regulation, i.e. with retroactive action, for influencing travelling speed, acceleration, or deceleration
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B1/00—Control systems of elevators in general
    • B66B1/34—Details, e.g. call counting devices, data transmission from car to control system, devices giving information to the control system
    • B66B1/3407—Setting or modification of parameters of the control system
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
    • B66B5/0006—Monitoring devices or performance analysers
    • B66B5/0012—Devices monitoring the users of the elevator system
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
    • B66B5/0006—Monitoring devices or performance analysers
    • B66B5/0018—Devices monitoring the operating condition of the elevator system
    • B66B5/0031—Devices monitoring the operating condition of the elevator system for safety reasons
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
    • B66B5/0006—Monitoring devices or performance analysers
    • B66B5/0037—Performance analysers
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
    • B66B5/02—Applications of checking, fault-correcting, or safety devices in elevators responsive to abnormal operating conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B2201/00—Aspects of control systems of elevators
    • B66B2201/40—Details of the change of control mode
    • B66B2201/46—Switches or switchgear
    • B66B2201/4607—Call registering systems
    • B66B2201/4638—Wherein the call is registered without making physical contact with the elevator system
    • B66B2201/4646—Wherein the call is registered without making physical contact with the elevator system using voice recognition
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B66—HOISTING; LIFTING; HAULING
    • B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
    • B66B2201/00—Aspects of control systems of elevators
    • B66B2201/40—Details of the change of control mode
    • B66B2201/46—Switches or switchgear
    • B66B2201/4607—Call registering systems
    • B66B2201/4653—Call registering systems wherein the call is registered using portable devices
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
    • Y02B50/00—Energy efficient technologies in elevators, escalators and moving walkways, e.g. energy saving or recuperation technologies

Abstract

The invention belongs to the technical field of elevators, and provides an intelligent elevator control system which comprises a core controller, a gesture recognition module, a voice recognition module, an acceleration running state module, a raspberry pie, a mobile terminal, a PC terminal, a development board, a human body infrared detection module, a temperature and humidity detection module and an oled display screen, wherein the gesture recognition module is used for recognizing the gesture of the user; according to the intelligent elevator control system based on the intelligent technology, intelligent control and management of the elevator are achieved, safety and service quality of the elevator are improved, the state of the elevator is monitored in real time, various parameters of the elevator are transmitted to professionals through the alarm system at the first time, basic judgment is carried out on operation parameters of the elevator, maintenance is carried out in time, safety problems are greatly reduced, the abnormal conditions can be monitored through the human body gesture recognition technology, alarm signals are automatically sent, and related personnel are informed of timely taking emergency measures, so that safety performance and service efficiency of the elevator can be effectively improved.

Description

Intelligent elevator control system
Technical Field
The invention belongs to the technical field of elevators, and particularly relates to an intelligent elevator control system.
Background
The intellectualization means that things have the capability of meeting various demands of human beings under the support of computer networks, big data, internet of things, artificial intelligence and other technologies, and along with the rapid development of modern communication technologies, computer network technologies and field bus control technologies, digitization, networking and informatization are gradually integrated into the lives of people. On the basis of the ever-increasing and improving of living levels and living conditions, there is a growing demand for intelligent elevator control systems, and higher demands are being made on the quality of life. The intelligent content is continuously integrated into new concepts, and the elevator is widely applied to life of people as a convenient tool.
In chinese patent publication No. CN108910632B, an intelligent safety elevator system is mentioned, comprising a background subsystem, an intelligent elevator and a number of elevator key panels; the intelligent elevator comprises an elevator body and an elevator controller, wherein the elevator body comprises a car, a stand column is vertically arranged in the middle of the car, and an installation cavity is formed in the stand column; 2 groups of power components are sequentially arranged in the mounting cavity from top to bottom, and each power component comprises a stepping motor, a mounting plate, a bottom plate, a turntable and a supporting rod, and the bottom plate and the mounting plate are both fixed on the inner wall of the mounting cavity; the turntable is rotationally connected to the bottom plate, the stepping motor is vertically downwards arranged on the mounting plate, and the output shaft is coaxially connected with the turntable; one end of the supporting rod is fixed on the side wall of the turntable, and the other end extends out of the upright post; the rod body of the support rod outside the upright post is provided with a movable partition board; the side wall of the upright post is also provided with a fixed baffle; an electric turntable is arranged at the bottom of the lift car.
However, the above-mentioned patent utilizes identification and intelligent control to carry out the regional division with the car of elevator, let different owner users get into the region of each allocation, realize relative isolation, can't accomplish the human gesture in the elevator and carry out alarm processing according to the circumstances like traditional elevator, simultaneously because elevator itself has complicated structure, simultaneously as the product of mechatronic, often receive many contingencies to take place in the course of using, lead to unexpected accident to take place, traditional elevator lacks the function of effectively monitoring the motion behavior in the car, in traditional elevator, need control elevator operation through pressing floor button or switch door button etc. however this kind of mode has some problems such as pressing wrong button, forget to press the button, inconvenient face complicated floor structure etc. in addition, along with the spread of virus in the space easily, cross infection problem in the public place becomes more and more outstanding.
Disclosure of Invention
In order to solve the technical problems, the invention provides an intelligent elevator control system which aims at solving the problems that the prior art lacks a function of effectively monitoring the movement behavior in a box body, can not control the elevator to run rapidly and intelligently and reduces cross infection in public places.
The intelligent elevator control system comprises a core controller, a gesture recognition module, a voice recognition module, an acceleration running state module, a raspberry group, a mobile terminal, a PC terminal, a development board, a human body infrared detection module, a temperature and humidity detection module and an oled display screen, and is characterized in that:
s1, selecting STM32F103C8T6 as a core controller, and sending an instruction for receiving data information;
s2, designing a frame difference method and a MobileNetV2 neural network model, and introducing an OpenMV module to realize a gesture recognition function by comparing and selecting the optimal;
s3, calculating the running state of the elevator through a large amount of data acquired by the sensor and setting a threshold value for alarming;
s4, training an LD3320 voice recognition module to perform voice recognition and setting a special wake-up word;
s5, designing an upper computer to display all parameters and information and carrying out feedback alarm.
Preferably, the motor is adopted to simulate the ascending and descending of the elevator, and the relay is adopted to simulate the elevator switch to test the feasibility.
Preferably, the frame difference method adopts a three-frame difference method to extract a target contour from a section of video image, the obtained difference image is subjected to binarization processing, and then the image is subjected to denoising, expansion corrosion, edge detection and gravity center calculation;
the denoising adopts a median filtering technology to remove a two-dimensional sliding template with a certain structure, the pixels in the plate are ordered according to the size of the pixel values, a monotonically ascending or descending two-dimensional data sequence is generated, and the two-dimensional median filtering output is as follows:
；
wherein ,，/>the processed image and the original image, respectively, are filtered, in a median filtering technique,representing a two-dimensional sliding template;
the expansion corrosion can expand and contract the size of the object, smooth the outside of the object, and makeRepresenting the input image +.>Represents structural elements (I)> and />Are respectively a function-> and />Is defined as:
；
the definition of gray scale expansion is:
；
the gray scale on operation is that the structural element B firstly carries out corrosion operation and then expansion operation on the gray scale image, namely:
；
the gray scale closing operation is that the structural element B firstly expands the gray scale image and then erodes the gray scale image, namely:
；
firstly, corroding the image once, and then expanding the image once to obtain an external outline object;
the edge detection adopts a Canny edge detection algorithm, a Canny operator finds an optimal edge detection, the optimal edge detection comprises the steps of identifying actual edges in an image as much as possible, the probability of missing detection of the actual edges and the probability of false detection of non-edges are both as small as possible, the detected edge points are nearest to the actual edge points, or the degree of deviation of the detected edges from the actual edges of the object due to noise influence is minimum, and the edge points detected by the operator correspond to the actual edge points one by one;
the center of gravity calculation is to calculate the center of gravity of the outside by using the coordinate value of each pixel point of the outside contour after extracting the outside contour of the person in the video, extract continuous frames of images to obtain a center of gravity value, and obtain an acceleration value according to the change displacement of the center of gravity value and a time formula.
Preferably, the acceleration value has a relation to the various attitudes of the human body in the elevator.
Preferably, the MobileNetV2 neural network model comprises the following steps:
s201, preparing a data set, and dividing the data set into a training set, a verification set and a test set, wherein the training set and the verification set are used for training and adjusting a model, the test set is used for evaluating the performance of the model, the training set accounts for about 80% of the total data set, and the verification set and the test set respectively account for 10%;
s202, loading a pre-training MobileNet V2 model, taking the MobileNet V2 model pre-trained on a large-scale image data set as a basic model, taking the pre-training model as a feature extractor through transfer learning, and then performing fine adjustment on the basis of the pre-training model;
s203, freezing the model, namely freezing the first layers of the model by utilizing the feature extraction capability of the pre-training model, training only the last layers of the model, wherein the first layers generally comprise general image features, the last layers comprise higher-level semantic information, reserving the first layers of the pre-training model, and adjusting the last layers of the model;
s204, adding a custom output layer, and adding a custom output layer after the last layer of the model for predicting the category of the image;
s205, training a model, training the model by using a training set, adjusting the model by using a verification set, and amplifying a data set by using a data enhancement technology during training;
s206, evaluating the model, and evaluating the trained model by using a test set;
s207, deploying a model, and performing classification prediction on a new image after model training and evaluation are completed;
the transfer learning utilizes a trained model to carry out fine tuning, reduces the time and the computing resources required by model training, and comprises the following steps:
s20201, selecting MobileNet V2 as a pre-training model;
s20202, removing the last layer of classifier of the pre-training model, and replacing the classifier with a new classifier;
s20203, freezing all parameters of the pre-training model, and training only a new classifier;
s20204, training with appropriate optimization algorithms and loss functions until satisfactory performance is achieved.
Preferably, the LD3320 voice recognition module comprises a signal receiving module and an information processing and interaction module, wherein the signal receiving module transmits an information message format specified by the single-chip microcomputer control module to the single-chip microcomputer processing module, and the information processing and interaction module controls the whole deviceThe stable operation of the development board uses the relation of interrupt coordination serial port transmission and DMA to decode the received voice information message, screens the information needed by the user from a large number of data messages, trains a binary classifier to judge whether the input audio contains specific wake-up words, and establishesFeature vector representing audio sample, ++>Indicating whether audio contains wake-up words-> wherein />Indicating that wake-up words are not included>Representing that the wake word is included;
is provided withIs training set, wherein->To train the number of samples, a classification function is learned to classify the input audio into two classes, namely:
；
for binary classification problems, a logistic regression model is used for prediction, and the logistic regression model is assumed to be:
；
wherein ,for model parameter vector, ++>Outputting the result for the model, +.>Is a row vector,/->A transpose operation representing a vector; thus (S)>The representation will->Vector sum->The vector is the result of the inner product multiplication, i.e. +.>Each element of the vector and->The corresponding elements of the vectors are multiplied and the products are added to obtain a scalar value, in this model,/>The result of (2) is passed to a sigmoid function for calculating the probability that the sample belongs to the positive case, learning model parameters such that the model outputs +.>As close as possible to the wake-up word->Thus, the model is trained using a method that minimizes the cross entropy loss function, namely:
；
in the case of a logistic regression, the data,representing a cost function for evaluating the gap between the model predictive result and the real label,/>Representing the number of samples, +.>Representing the i-th sample, +.>The value of (2) is in the range of 1 to m, (-)>A parameter vector representing the model, wherein->Is a column vector,/->Representing the prediction result of the logistic regression model on the ith sample, wherein +.>Is a line vector representing the eigenvector of the ith sample, < >>The true label representing the i-th sample, takes a value of 0 or 1,the meaning of (a) is that for all samples, the difference between the model prediction result and the real label is calculated and then averaged, wherein when the samples belong to positive examples, the value of the cost function depends on the prediction probability of the model alignment example; when the sample belongs to the negative example, the value of the cost function depends on the prediction probability of the model on the negative example; />The smaller the value of (2), the smaller the gap between the predicted result of the model and the real label, the modelThe better the performance of (c).
Preferably, the threshold value monitors the running state parameters of the elevator in real time through a sensor and acquires data, the collected data is preprocessed, and a plurality of original independent variables are obtainedConsidered as a time series, and subjected to a discrete wavelet transform DWT,
the wavelet coefficients and scale coefficients are calculated by the following formula:
；
；
wherein ,，/>represents the scale factor and wavelet factor of the j-th layer, respectively,>，/>the low-pass and high-pass filter coefficients in the wavelet basis function, respectively, < + >>Is an offset, +.>The number of layers is the number of decomposition, a high-frequency wavelet coefficient is selected, and the common characteristics in a plurality of independent variables are highlighted;
for a set of training data,, wherein />Representing sample feature vectors, ++>Representing class labels, the goal of the support vector machine SVM is to find a hyperplane, separate samples of different classes, and maximize the distance of the nearest sample point to the hyperplane, then the hyperplane can be represented as:
；
wherein Representing the two-vector inner product, assuming that the distances from the nearest positive example point and the nearest negative example point to the hyperplane are respectivelyThe constraint is expressed as:
；
；
the objective function is expressed as a minimum and the SVM as the following optimization problem:
；
；
the optimization problem is solved using a gorilla optimization algorithm.
Preferably, the S3 further comprises a built-in six-axis gyroscope, safety recognition and detection are continuously carried out during the operation of the elevator, an alarm is automatically sent to a remote terminal of the elevator when an abnormal operation track of the elevator is detected, an external personnel recognition module is additionally arranged at an upper control terminal and a lower control terminal of the elevator, the waiting condition of a user is dynamically monitored through a human body infrared detection module, and an external elevator taking request is automatically ended when the abnormal condition is found.
Preferably, the development board adopts Arduino UNO Rev3, and is used for receiving related data information of the acceleration running state module, the human body infrared detection module and the temperature and humidity detection module and transmitting the data information to the raspberry group, and the raspberry group transmits the data information to the mobile terminal and the PC terminal through an MQTT communication mode, and meanwhile, the data information is dynamically displayed on the oled display screen.
Compared with the prior art, the invention has the following beneficial effects:
according to the elevator control system based on the intelligent technology, multiple technologies of deep learning and computer vision are adopted, so that the voice recognition and control function, the automatic sterilization function, the gesture recognition function and the operation detection and alarm function are realized, the intelligent control and management of the elevator are realized, and the safety and the service quality of the elevator are improved.
Drawings
FIG. 1 is a schematic view of an overall frame of the present invention;
FIG. 2 is a detailed frame schematic of the present invention;
FIG. 3 is a schematic diagram of the three frame difference method, the dilation-erosion process and the edge detection process of the present invention;
FIG. 4 is a schematic diagram of the relationship between various poses and accelerations of the present invention;
fig. 5 is an internet of things interface display diagram of the present invention.
Detailed Description
Embodiments of the present invention are described in further detail below with reference to the accompanying drawings and examples. The following examples are illustrative of the invention but are not intended to limit the scope of the invention.
As shown in fig. 1 to 5:
embodiment one: the invention provides an intelligent elevator control system, which comprises a core controller, a gesture recognition module, a voice recognition module, an acceleration running state module, a raspberry group, a mobile terminal, a PC terminal, a development board, a human body infrared detection module, a temperature and humidity detection module and an oled display screen, and is characterized in that the intelligent elevator control system comprises the following steps:
s1, selecting STM32F103C8T6 as a core controller, and sending an instruction for receiving data information;
s2, designing a frame difference method and a MobileNetV2 neural network model, and introducing an OpenMV module to realize a gesture recognition function by comparing and selecting the optimal;
s3, calculating the running state of the elevator through a large amount of data acquired by the sensor and setting a threshold value for alarming;
s4, training an LD3320 voice recognition module to perform voice recognition and setting a special wake-up word;
s5, designing an upper computer to display all parameters and information and carrying out feedback alarm.
Specifically, a motor is adopted to simulate the ascending and descending of an elevator, and a relay simulates an elevator switch to test feasibility; through the ascending of motor forward simulation elevator, the descending of motor reverse simulation elevator connects the circuit of elevator door to electromagnetic relay's center pin and normally open pin, and when relay was triggered, the center pin will be connected to normally close pin to open the elevator door, when need close the elevator door, stop triggering relay, the center pin will return to normal position to break off the circuit of elevator door, accomplish the closure of elevator door.
Specifically, the frame difference method adopts a three-frame difference method to extract a target contour from a section of video image, the obtained difference image is subjected to binarization processing, and then the image is subjected to denoising, expansion corrosion, edge detection and gravity center calculation; when abnormal object motion occurs in a monitored scene, obvious difference occurs between frames, two frames are subtracted to obtain the absolute value of the brightness difference of two frames, a three-frame difference method is adopted to better extract a target contour, a threshold value is set, the pixel value is white above the threshold value, and all other pixels are black;
denoising adopts a median filtering technology to remove a two-dimensional sliding template with a certain structure, sorts pixels in a plate according to the size of pixel values, generates a monotonically ascending (or descending) two-dimensional data sequence, and outputs the two-dimensional median filtering as follows:
；
wherein ,，/>the processed image and the original image, respectively, are filtered, in a median filtering technique,representing a two-dimensional sliding template;
the expansion corrosion can expand and contract the size of the object, smooth the outside of the object, and makeRepresenting the input image +.>Represents structural elements (I)> and />Are respectively a function-> and />Is defined as:
；
the definition of gray scale expansion is:
；
the gray scale on operation is that the structural element B firstly carries out corrosion operation and then expansion operation on the gray scale image, namely:
；
the gray scale closing operation is that the structural element B firstly expands the gray scale image and then erodes the gray scale image, namely:
；
firstly, corroding the image once, and then expanding the image once to obtain an external outline object;
the edge detection adopts a Canny edge detection algorithm, a Canny operator finds an optimal edge detection, the optimal edge detection comprises the steps of identifying actual edges in an image as much as possible, the probability of missing detection of the actual edges and the probability of false detection of non-edges are both as small as possible, the detected edge points are nearest to the actual edge points, or the degree of deviation of the detected edges from the actual edges of the object due to noise influence is minimum, and the edge points detected by the operator correspond to the actual edge points one by one;
the gravity center calculation is to calculate the gravity center of the outside by using the coordinate value of each pixel point of the outside contour after extracting the outside contour of the person in the video;
and (5) calculating the center of gravity:
；
setting the contour line asIs arranged atThe upper point isSince the gray values inside the contour are the same, they can be regarded as being of equal density, atGet up a bitThen alongTaking another pointThe area surrounded by the two-point connecting line and the outline is(depending on the number of pixels enclosed),the area of the whole contour (the number of pixels surrounded by the whole contour) is determined according to the above equationObtaining line segmentsAt the same time take a bit (different fromAnd) The same method is used for solving the line segments＇' find the straight line where the line segment is locatedAndthe expression:
；
the intersection of the two lines is the center of gravity；
Extracting continuous frames of images to obtain a gravity center value, and obtaining an acceleration value according to the gravity center value change displacement and a time formula:
；
for acceleration, inThe gravity center change value in time isIn the followingThe gravity center change value in time is，And/2, representing the average value of the gravity center change values in the two time periods, and thus, estimating the magnitude of the acceleration by calculating the average value of the gravity center change values and the time length of the two time periods.
Specifically, the acceleration value has a relation with various postures:
；
；
；
representing the number of times that the test is normal when bending or squatting,the number of falls detected when the watch is bent or squat down,representing the number of times that normal walking is detected as normal,representing the case where normal walking detection is a fall,representative is the number of detections in a fall,representing the number of times a fall has occurred that is not detected, wherein,indicating the probability of being detected as normal when bending or squatting;the probability of being detected as normal during normal walking is represented;representing the probability of being detected as a fall when it is.
Specifically, the mobilenet v2 neural network model includes the following steps:
s201, preparing a data set, and dividing the data set into a training set, a verification set and a test set, wherein the training set and the verification set are used for training and adjusting a model, the test set is used for evaluating the performance of the model, the training set accounts for about 80% of the total data set, and the verification set and the test set respectively account for 20%;
s202, loading a pre-training MobileNet V2 model, taking the MobileNet V2 model pre-trained on a large-scale image data set as a basic model, taking the pre-training model as a feature extractor through transfer learning, and then performing fine adjustment on the basis of the pre-training model;
s203, freezing the model, namely freezing the first layers of the model by utilizing the feature extraction capability of the pre-training model, training only the last layers of the model, wherein the first layers generally comprise general image features, the last layers comprise higher-level semantic information, reserving the first layers of the pre-training model, and adjusting the last layers of the model;
s204, adding a custom output layer, and adding a custom output layer after the last layer of the model for predicting the category of the image;
s205, training a model, training the model by using a training set, adjusting the model by using a verification set, and amplifying a data set by using a data enhancement technology during training;
s206, evaluating the model, and evaluating the trained model by using a test set;
s207, deploying a model, and performing classification prediction on a new image after model training and evaluation are completed;
the transfer learning utilizes the trained model to carry out fine tuning, reduces the time and calculation resources required by model training, and comprises the following steps:
s20201, selecting MobileNet V2 as a pre-training model;
s20202, removing the last layer of classifier of the pre-training model, and replacing the classifier with a new classifier;
s20203, freezing all parameters of the pre-training model, and training only a new classifier;
s20204, training by adopting a proper optimization algorithm and a loss function until satisfactory performance is achieved;
the mathematical derivation of mobilenet v2 is as follows:
assuming that the size of the input picture is h×w×c, where H represents the height, W represents the width, C represents the number of channels, for example, the number of channels of the RGB image is 3, and the model input picture is output as a feature map with size h×w×c after a series of operations such as convolution, normalization, and activation functions, where H represents the height of the feature map, W represents the width of the feature map, and C represents the number of channels of the feature map, that is, the feature number;
MobileNetV2 uses mainly depth separable convolution, which consists of two parts: depth convolution and point-by-point convolution; the deep convolution represents a convolution operation performed on each channel, so as to obtain C feature images with the size of H multiplied by W, wherein the pixel value of each feature image is as follows:
；
wherein ,the weights of the convolution kernel are represented,the kth channel position representing the input signaturePixel values of (2); the point-by-point convolution means that convolution operation is performed on each pixel point, so as to obtain C' feature graphs with the size of h multiplied by w, wherein the pixel value of each feature graph is as follows:
；
wherein ,the weights of the convolution kernel are represented,the kth channel position representing the feature map obtained by depth convolutionPixel values of (2);
according to the elevator simulation system, the function of simulating an elevator is realized through the relay and the motor, the target motion state in the elevator is accurately detected through the frame difference method machine vision and the MobileNet V2 neural network model, the stability and the safety of the elevator are improved, meanwhile, the model is imported into the OpenMV model, the efficient calculation and the low-power-consumption operation of the neural network are realized, the elevator simulation system is suitable for application scenes needing long-time operation, the OpenMV model has the capability of rapid iteration, the model performance is optimized through modes of fine adjustment model parameters, data enhancement and the like, the test and the verification are carried out in a short time, and meanwhile, the OLED screen is used for dynamically displaying floor information identified in an elevator control system.
Embodiment two: the embodiment is basically the same as the previous embodiment, in that the LD3320 voice recognition module includes a signal receiving module, an information processing and interaction module, the signal receiving module transmits an information message format specified by the single chip microcomputer control module to the single chip microcomputer processing module, the information processing and interaction module controls the stable operation of the whole development board, decodes the received voice information message by using the relation of interrupt coordination serial port transmission and DMA, screens out information needed by a user from a large number of data messages, trains a binary classifier to judge whether the input audio contains specific wake-up words, and establishesFeature vector representing audio sample, ++>Indicating whether audio contains wake-up words-> wherein />Indicating that wake-up words are not included>Representing that the wake word is included;
is provided withIs training set, wherein->To train the number of samples, a classification function is learned to classify the input audio into two classes, namely:
；
for binary classification problems, a logistic regression model is used for prediction, and the logistic regression model is assumed to be:
；
wherein ,for model parameter vector, ++>Outputting the result for the model, +.>Is a row vector,/->A transpose operation representing a vector; thus (S)>The representation will->Vector sum->The vector is the result of the inner product multiplication, i.e. +.>Each element of the vector and->The corresponding elements of the vectors are multiplied and the products are added to obtain a scalar value, in this model,/>The result of (2) is passed to a sigmoid function for calculating the probability that the sample belongs to the positive case, learning model parameters such that the model outputs +.>As close as possible to the wake-up word->Thus, the model is trained using a method that minimizes the cross entropy loss function, namely:
；
in the case of a logistic regression, the data,representing a cost function for evaluating the gap between the model predictive result and the real label,/>Representing the number of samples, +.>Representing the i-th sample, +.>The value of (2) is in the range of 1 to m, (-)>A parameter vector representing the model, wherein->Is a column vector,/->Representing the predicted junction of the logistic regression model to the ith sampleFruit of which->Is a line vector representing the eigenvector of the ith sample, < >>The true label representing the i-th sample, takes a value of 0 or 1,the meaning of (a) is that for all samples, the difference between the model prediction result and the real label is calculated and then averaged, wherein when the samples belong to positive examples, the value of the cost function depends on the prediction probability of the model alignment example; when the sample belongs to the negative example, the value of the cost function depends on the prediction probability of the model on the negative example; />The smaller the value of (2), the smaller the gap between the prediction result of the model and the real label, the better the performance of the model, and the model is used for predicting whether the new audio sample contains wake-up words; outputting the model result +.>Comparing the audio input with a preset threshold value, judging whether the audio input contains a wake-up word according to the threshold value, and triggering corresponding operation in time; preprocessing an audio sample to improve the accuracy of subsequent training, converting the preprocessed audio sample into a feature vector for training a classifier, training the classifier by using a machine learning algorithm, marking a wake-up word as a positive example in the training process, marking the rest as a negative example, performing operations such as super-parameter adjustment and cross verification according to requirements to improve the performance of the classifier, testing and evaluating the trained classifier by using a test set, deploying the trained wake-up word model into an LD3320 voice recognition module, and performing real-time running test.
Specifically, the threshold value monitors elevator running state parameters in real time through a sensor and acquires data, the collected data is preprocessed, and a plurality of original independent variables are obtainedRegarding as a time sequence, performing Discrete Wavelet Transform (DWT) on the time sequence, writing a program by using programming languages such as Python, acquiring real-time data from an elevator parameter sensor, processing and visualizing the real-time data, writing corresponding control logic and a user interface, connecting a raspberry group into an elevator system, performing real-time test and optimization, deploying a written elevator parameter upper computer application program into the raspberry group, setting the application program to be started up for self-starting, ensuring stable operation of the program, and timely processing and displaying elevator parameter data.
The wavelet coefficients and scale coefficients are calculated by the following formula:
；/>
；
wherein ,，/>represents the scale factor and wavelet factor of the j-th layer, respectively,>，/>the low-pass and high-pass filter coefficients in the wavelet basis function, respectively, < + >>Is an offset, +.>The number of layers is the number of decomposition, a high-frequency wavelet coefficient is selected, and the common characteristics in a plurality of independent variables are highlighted;
for a set of training data,, wherein />Representing sample feature vectors, ++>Representing class labels, the goal of the support vector machine SVM is to find a hyperplane, separate samples of different classes, and maximize the distance of the nearest sample point to the hyperplane, then the hyperplane can be represented as:
；
wherein Representing the two-vector inner product, assuming that the distances from the nearest positive example point and the nearest negative example point to the hyperplane are respectivelyThe constraint is expressed as:
；
；
the objective function is expressed as a minimum and the SVM as the following optimization problem:
；
；
the optimization problem is solved using a gorilla optimization algorithm.
Specifically, S3 also comprises a built-in six-axis gyroscope which can continuously perform safety identification and detection during the operation of the elevator, automatically send an alarm to a remote end of the elevator when detecting that an abnormal operation track occurs to the elevator, and an external personnel identification module is additionally arranged at an upper control end and a lower control end of the elevator, dynamically monitors the waiting condition of a user through a human body infrared detection module, and automatically ends an external elevator taking request when detecting the abnormal condition; an ADXL355 acceleration sensor in the module obtains a vertical acceleration value in the vertical direction of the elevator car at each momentAnd a horizontal acceleration value in the horizontal direction of the moving car doorAnd store according to time sequence valueAndthe method comprises the steps of carrying out a first treatment on the surface of the Based onObtaining vertical travel speed of elevator car in vertical direction at each momentAnd pass throughAndcalculating the vertical displacement distance of the elevator car in the vertical direction at each momentAnd store according to time sequence valueAndthe method comprises the steps of carrying out a first treatment on the surface of the Based onAcquiring horizontal running speed of a moving car door in the horizontal direction at each momentAnd store according to time sequence valueThe method comprises the steps of carrying out a first treatment on the surface of the Based on、、、Andjudging the running state of the elevator car at each moment with the running state reference value, and comparing the running state of the elevator car with each running state reference value based on the parameter value; and judging the running state of the elevator car at each moment based on the comparison result, judging whether the running state of the elevator car triggers alarm processing, outputting alarm information through an alarm device of the elevator car if the alarm processing is triggered, and simultaneously transmitting the information to elevator maintenance personnel in real time to perform unscheduled repair.
From the above, the operation such as going up and down, opening and closing the door of elevator is realized through voice control, the convenience of use and comfort level of user have been improved, training wake-up word adopts multiple mode to collect voice data, including the voice data of different crowds, different environment and different accents, with the wider application scenario of coverage, will wake-up word detection task and other voice task joint training, in order to improve the performance and the user experience of whole speech recognition system, the running state and the fault information of system still can real-time supervision elevator, and in time show each item information through the host computer, the security and the reliability of elevator have been improved.
Embodiment III: the difference between the embodiment and the previous embodiment is that the development board adopts Arduino UNO Rev3 for receiving the related data information of the acceleration running state module, the human body infrared detection module and the temperature and humidity detection module and transmitting the data information to the raspberry group, and the raspberry group transmits the data information to the mobile terminal and the PC terminal through an MQTT communication mode and simultaneously dynamically displays the data information on an oled display screen; when a target leaves the elevator, the HC-SR501 human body infrared identification module sensor transmits signals to the processor, and the nozzle is controlled to spray disinfectant for disinfection through temperature and humidity information provided by the DHT11 sensor.
Meanwhile, the data network transmission is carried out between the mobile terminal and the PC terminal and the server by adopting an MQTT communication mode, the temperature, humidity and acceleration theme information is subscribed through the mobile terminal and the PC terminal, then Esp and 8266 are used for publishing the data detected by the elevator system to the temperature, humidity and theme, the data are transmitted to the Internet of things server by adopting an MQTT communication protocol, and finally the Internet of things server can send the temperature, humidity and theme information to the mobile terminal and the PC terminal at any moment when the data are needed.
From the above, the ultraviolet sterilization technology can be utilized to automatically sterilize the inside of the elevator, so that the sanitary safety of the elevator is improved, the risk of cross infection is greatly reduced, and a user receives temperature and humidity information through an efficient MQTT communication mode.
Working principle: through the elevator control system based on intelligent technology, the multiple technologies of deep learning and computer vision are adopted, the voice recognition and control function, the automatic sterilization function, the gesture recognition function and the operation detection and alarm function are realized, the intelligent control and management of the elevator are realized, the safety and service quality of the elevator are improved, the elevator state is monitored in real time, various parameters of the elevator can be transmitted to professionals through an alarm system at the first time, the elevator operation parameters are basically judged, the elevator is maintained in time, the safety problem is greatly reduced, the abnormal conditions can be monitored by the human gesture recognition technology, alarm signals are automatically sent, and related personnel are informed to take emergency measures in time, so that the safety performance and the service efficiency of the elevator can be effectively improved.
While embodiments of the present invention have been shown and described above for purposes of illustration and description, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives, and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims (7)

1. The intelligent elevator control system is characterized by comprising a core controller, a gesture recognition module, a voice recognition module, an acceleration running state module, a raspberry group, a mobile terminal, a PC terminal, a development board, a human body infrared detection module, a temperature and humidity detection module and an oled display screen, and is characterized by comprising the following steps:
s1, selecting STM32F103C8T6 as a core controller, and sending an instruction for receiving data information;
s2, designing a frame difference method and a MobileNetV2 neural network model, and introducing an OpenMV module to realize a gesture recognition function by comparing and selecting the optimal;
s3, calculating the running state of the elevator through a large amount of data acquired by the sensor and setting a threshold value for alarming;
s4, training an LD3320 voice recognition module to perform voice recognition and setting a special wake-up word;
s5, designing an upper computer to display all parameters and information and perform feedback alarm;
wherein the mobilenet v2 neural network model comprises the steps of:
s201, preparing a data set, and dividing the data set into a training set, a verification set and a test set, wherein the training set and the verification set are used for training and adjusting a model, the test set is used for evaluating the performance of the model, the training set accounts for about 80% of the total data set, and the verification set and the test set respectively account for 10%;
s202, loading a pre-training MobileNet V2 model, taking the MobileNet V2 model pre-trained on a large-scale image data set as a basic model, taking the pre-training model as a feature extractor through transfer learning, and then performing fine adjustment on the basis of the pre-training model;
s203, freezing the model, namely freezing the first layers of the model by utilizing the feature extraction capability of the pre-training model, training only the last layers of the model, wherein the first layers generally comprise general image features, the last layers comprise higher-level semantic information, reserving the first layers of the pre-training model, and adjusting the last layers of the model;
s204, adding a custom output layer, and adding a custom output layer after the last layer of the model for predicting the category of the image;
s205, training a model, training the model by using a training set, adjusting the model by using a verification set, and amplifying a data set by using a data enhancement technology during training;
s206, evaluating the model, and evaluating the trained model by using a test set;
s207, deploying a model, and performing classification prediction on a new image after model training and evaluation are completed;
the transfer learning utilizes a trained model to carry out fine tuning, reduces the time and the computing resources required by model training, and comprises the following steps:
s20201, selecting MobileNet V2 as a pre-training model;
s20202, removing the last layer of classifier of the pre-training model, and replacing the classifier with a new classifier;
s20203, freezing all parameters of the pre-training model, and training only a new classifier;
s20204, training by adopting a proper optimization algorithm and a loss function until satisfactory performance is achieved; the LD3320 voice recognition module comprises a signal receiving module and an information processing and interacting module, wherein the signal receiving module transmits an information message format regulated by the singlechip control module to the singlechip processing module, the information processing and interacting module controls the stable operation of the whole development board, decodes the received voice information message by using the relation of interrupt coordination serial port transmission and DMA, screens information needed by a user from a large number of data messages, and trains a binary classifier to judgeIf the input audio contains specific wake-up words, it is setFeature vector representing audio sample, ++>Indicating whether audio contains wake-up words-> wherein />Indicating that wake-up words are not included>Representing that the wake word is included;
is provided withIs training set, wherein->To train the number of samples, a classification function is learned to classify the input audio into two classes, namely:
；
for binary classification problems, a logistic regression model is used for prediction, and the logistic regression model is assumed to be:
；
wherein ,for model parameter vector, ++>Outputting the result for the model, +.>Is a row vector,/->A transpose operation representing a vector; thus (S)>The representation will->Vector sum->The vector is the result of the inner product multiplication, i.e. +.>Each element of the vector and->The corresponding elements of the vectors are multiplied and the products are added to obtain a scalar value, in this model,/>The result of (2) is transferred to a sigmoid function for calculating the probability of the sample belonging to the positive example, and learning model parameters so as to make the model outputAs close as possible to the wake-up word->Thus, the model is trained using a method that minimizes the cross entropy loss function, namely:
；
in the case of a logistic regression, the data,representing a cost function for evaluating the gap between the model predictive result and the real label,representing the number of samples, +.>Representing the i-th sample, +.>The value of (2) is in the range of 1 to m, (-)>A parameter vector representing the model, wherein->Is a column vector,/->Representing the prediction result of the logistic regression model on the ith sample, wherein +.>Is a line vector representing the eigenvector of the ith sample, < >>The true label representing the ith sample is 0 or 1,/for the value>The meaning of (a) is that for all samples, the difference between the model prediction result and the real label is calculated and then averaged, wherein when the samples belong to positive examples, the value of the cost function depends on the prediction probability of the model alignment example; when the sample belongs to the negative example, the value of the cost function depends on the model versus the negative examplePredicting probability; />The smaller the value of (c) is, the smaller the gap between the predicted result of the model and the real label is, and the better the performance of the model is.
2. An intelligent elevator control system as set forth in claim 1, wherein: the motor is adopted to simulate the elevator to go up and down, and the relay is adopted to simulate the elevator switch to test the feasibility.
3. An intelligent elevator control system as set forth in claim 1, wherein: the frame difference method adopts a three-frame difference method to extract a target contour from a section of video image, the obtained differential image is subjected to binarization processing, and then the image is subjected to denoising, expansion corrosion, edge detection and gravity center calculation;
the denoising adopts a median filtering technology to remove a two-dimensional sliding template with a certain structure, the pixels in the plate are ordered according to the size of the pixel values, a monotonically ascending or descending two-dimensional data sequence is generated, and the two-dimensional median filtering output is as follows:
；
wherein ,，/>respectively processed image and original image, in median filtering technique +.>Representing a two-dimensional sliding template;
the expansion corrosion can expand and contract the size of the object, smooth the outside of the object, and makeRepresenting the input image and,represents structural elements (I)> and />Are respectively a function-> and />Is defined as:
；
the definition of gray scale expansion is:
；
the gray scale on operation is that the structural element B firstly carries out corrosion operation and then expansion operation on the gray scale image, namely:
；
the gray scale closing operation is that the structural element B firstly expands the gray scale image and then erodes the gray scale image, namely:
；
firstly, corroding the image once, and then expanding the image once to obtain an external outline object;
the edge detection adopts a Canny edge detection algorithm, a Canny operator finds an optimal edge detection, the optimal edge detection comprises the steps of identifying actual edges in an image as much as possible, the probability of missing detection of the actual edges and the probability of false detection of non-edges are both as small as possible, the detected edge points are nearest to the actual edge points, or the degree of deviation of the detected edges from the actual edges of the object due to noise influence is minimum, and the edge points detected by the operator correspond to the actual edge points one by one;
the center of gravity calculation is to calculate the center of gravity of the outside by using the coordinate value of each pixel point of the outside contour after extracting the outside contour of the person in the video, extract continuous frames of images to obtain a center of gravity value, and obtain an acceleration value according to the change displacement of the center of gravity value and a time formula.
4. An intelligent elevator control system as set forth in claim 3, wherein: the acceleration value has a relation with various attitudes of a human body in the elevator.
5. An intelligent elevator control system as set forth in claim 1, wherein: the threshold value monitors elevator running state parameters in real time through a sensor and acquires data, the collected data are preprocessed, and a plurality of original independent variables are obtainedConsidered as a time series, and subjected to a discrete wavelet transform DWT,
the wavelet coefficients and scale coefficients are calculated by the following formula:
；
；
wherein ,，/>represents the scale factor and wavelet factor of the j-th layer, respectively,>，/>the low-pass and high-pass filter coefficients in the wavelet basis function, respectively, < + >>Is an offset, +.>The number of layers is the number of decomposition, a high-frequency wavelet coefficient is selected, and the common characteristics in a plurality of independent variables are highlighted;
for a set of training data,, wherein />Representing sample feature vectors, ++>Representing class labels, the goal of the support vector machine SVM is to find a hyperplane, separate samples of different classes, and maximize the distance of the nearest sample point to the hyperplane, then the hyperplane can be represented as:
；
wherein Representing two vectors inThe product is obtained by assuming that the distances from the nearest positive example point and the nearest negative example point to the hyperplane are respectivelyThe constraint is expressed as:
；
；
the objective function is expressed as a minimum and the SVM as the following optimization problem:
；
；
the optimization problem is solved using a gorilla optimization algorithm.
6. An intelligent elevator control system as set forth in claim 1, wherein: s3 still include emergent alarm module, its six built-in gyroscopes will be in the elevator operation period incessant carry out safety recognition and detection, will send the alarm to the elevator remote end automatically when detecting that the elevator appears unusual movement track, outside personnel identification module installs control end about the elevator additional, through human infrared detection module dynamic monitoring user' S wait condition, discover that unusual condition is automatic to be ended external taking advantage of the ladder request.
7. An intelligent elevator control system as set forth in claim 1, wherein: the development board adopts Arduino UNO Rev3 and is used for receiving related data information of the acceleration running state module, the human body infrared detection module and the temperature and humidity detection module and transmitting the data information to the raspberry group, and the raspberry group transmits the data information to the mobile terminal and the PC terminal through an MQTT communication mode and dynamically displays the data information on the oled display screen.
CN202310728163.0A 2023-06-20 2023-06-20 Intelligent elevator control system Active CN116443682B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310728163.0A CN116443682B (en) 2023-06-20 2023-06-20 Intelligent elevator control system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310728163.0A CN116443682B (en) 2023-06-20 2023-06-20 Intelligent elevator control system

Publications (2)

Publication Number Publication Date
CN116443682A CN116443682A (en) 2023-07-18
CN116443682B true CN116443682B (en) 2023-08-25

Family

ID=87120600

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310728163.0A Active CN116443682B (en) 2023-06-20 2023-06-20 Intelligent elevator control system

Country Status (1)

Country Link
CN (1) CN116443682B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117556520B (en) * 2023-12-29 2024-03-15 南京瑞永城市更新研究院有限公司 VR collaborative processing system of existing house additional installation elevator

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB9518069D0 (en) * 1994-09-20 1995-11-08 Hitachi Ltd Method and apparatus for controlling elevators
JP2000351547A (en) * 1999-06-11 2000-12-19 Toshiba Corp Crime preventing operation device of elevator
CN109345507A (en) * 2018-08-24 2019-02-15 河海大学 A kind of dam image crack detection method based on transfer learning
CN110989689A (en) * 2019-12-16 2020-04-10 兰州理工大学 Robot system based on automatic tracking
CN214141091U (en) * 2020-11-26 2021-09-07 兰州理工大学 Intelligent elevator management system based on face and voice interaction technology
CN114333821A (en) * 2021-12-30 2022-04-12 山东声智物联科技有限公司 Elevator control method, device, electronic equipment, storage medium and product
CN114873387A (en) * 2022-04-12 2022-08-09 武汉理工大学 Elevator energy-saving dispatching system and method based on reinforcement learning algorithm

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB9518069D0 (en) * 1994-09-20 1995-11-08 Hitachi Ltd Method and apparatus for controlling elevators
JPH0885682A (en) * 1994-09-20 1996-04-02 Hitachi Ltd Operational control of elevator and its device
JP2000351547A (en) * 1999-06-11 2000-12-19 Toshiba Corp Crime preventing operation device of elevator
CN109345507A (en) * 2018-08-24 2019-02-15 河海大学 A kind of dam image crack detection method based on transfer learning
CN110989689A (en) * 2019-12-16 2020-04-10 兰州理工大学 Robot system based on automatic tracking
CN214141091U (en) * 2020-11-26 2021-09-07 兰州理工大学 Intelligent elevator management system based on face and voice interaction technology
CN114333821A (en) * 2021-12-30 2022-04-12 山东声智物联科技有限公司 Elevator control method, device, electronic equipment, storage medium and product
CN114873387A (en) * 2022-04-12 2022-08-09 武汉理工大学 Elevator energy-saving dispatching system and method based on reinforcement learning algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于卷积神经网络的维吾尔语语音识别;梁玉龙;屈丹;李真;张文林;;信息工程大学学报(第01期);第44-50页 *

Also Published As

Publication number Publication date
CN116443682A (en) 2023-07-18

Similar Documents

Publication Publication Date Title
CN109522793B (en) Method for detecting and identifying abnormal behaviors of multiple persons based on machine vision
CN107679471B (en) Indoor personnel air post detection method based on video monitoring platform
CN106241584A (en) A kind of intelligent video monitoring system based on staircase safety and method
US20220185625A1 (en) Camera-based sensing devices for performing offline machine learning inference and computer vision
CN116443682B (en) Intelligent elevator control system
CN105100724A (en) Remote and safe intelligent household monitoring method and device based on visual analysis
CN202257856U (en) Driver fatigue-driving monitoring device
CN113424221A (en) Model generation device, method, program, and prediction device
US11403879B2 (en) Method and apparatus for child state analysis, vehicle, electronic device, and storage medium
US11935297B2 (en) Item monitoring for doorbell cameras
JP6587268B1 (en) Platform risk determination program and system
CN111223263A (en) Full-automatic comprehensive fire early warning response system
CN110909672A (en) Smoking action recognition method based on double-current convolutional neural network and SVM
CN108776452B (en) Special equipment field maintenance monitoring method and system
CN115546899A (en) Examination room abnormal behavior analysis method, system and terminal based on deep learning
CN113095160A (en) Power system personnel safety behavior identification method and system based on artificial intelligence and 5G
CN110705413B (en) Emotion prediction method and system based on sight direction and LSTM neural network
Farzad et al. Recognition and classification of human behavior in Intelligent surveillance systems using Hidden Markov Model
CN112926364A (en) Head posture recognition method and system, automobile data recorder and intelligent cabin
CN109951866B (en) People flow monitoring method based on hidden Markov model
CN215520976U (en) Cable tunnel personnel are detained monitoring system
Acampora et al. Interoperable services based on activity monitoring in Ambient Assisted Living environments
US20240020963A1 (en) Object embedding learning
CN116894978B (en) Online examination anti-cheating system integrating facial emotion and behavior multi-characteristics
KR102480150B1 (en) Disease occurrence prediction system and method using three-dimensional multichannel data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant