CN108675071B

CN108675071B - Cloud cooperative intelligent chip based on artificial neural network processor

Info

Publication number: CN108675071B
Application number: CN201810413695.4A
Authority: CN
Inventors: 高钰峰; 陈云霁
Original assignee: Institute of Computing Technology of CAS
Current assignee: Institute of Computing Technology of CAS
Priority date: 2018-05-03
Filing date: 2018-05-03
Publication date: 2020-01-17
Anticipated expiration: 2038-05-03
Also published as: CN108675071A

Abstract

The present disclosure provides an artificial intelligence elevator dispatching system, including: the system comprises a plurality of cameras based on artificial intelligence, a plurality of sensors and a plurality of sensors, wherein the cameras are configured to be installed on different floors of a building and used for capturing waiting personnel outside the elevator and outputting the number of the counted personnel; the artificial intelligent elevator dispatching equipment is used for responding to the user request of calling floors, receiving the user request data of at least one floor and determining an elevator dispatching scheme. The number of people in the shot picture and/or video can be analyzed at the camera head end, basis is provided for later analysis, and the video does not need to be transmitted to a background to be processed.

Description

Cloud cooperative intelligent chip based on artificial neural network processor

Technical Field

The disclosure relates to the technical field of information processing, in particular to an artificial intelligence elevator dispatching system.

Background

The existing camera technology that relates to includes: the method adopts various optical and electric signal inputs to be converted into electric signals to be stored in a storage medium or directly stored as an entity storage medium, and the information of the storage medium is read into computer software to carry out image recognition and output a required signal result. The existing intelligent camera has the technical problems that firstly, the image recognition can not be directly carried out in the camera, the whole system needs wired or wireless connection for data transmission, the system is huge, and the execution efficiency is low; moreover, image recognition is mainly performed through software, power consumption is large, efficiency is low, and good real-time performance cannot be achieved.

In the existing elevator dispatching system, a forward hijacking (referring to a LOOK (LOOK-up) dispatching algorithm of hard disk dispatching, and having the academic name of an operating system Scan algorithm) method, a near-corresponding (nearest elevator corresponding service) method or a neural network balance dispatching method without considering the number of people waiting for an elevator are adopted.

The problem of current elevator dispatch system technique lies in: firstly, scheduling optimization is difficult to realize during group control elevator scheduling, so that elevator car resources can be reasonably and fully utilized; secondly, the optimization algorithm is difficult to realize, so that the average response time of the elevator taking request is shortest; in addition, the elevator dispatching mechanism is relatively fixed, real-time optimization cannot be achieved, and learning ability and adaptability under practical application scenes are lacked; and in addition, real-time people flow data cannot be detected, so that a proper elevator resource scheduling mechanism is adjusted.

Disclosure of Invention

Technical problem to be solved

In view of the above, it is an object of the present disclosure to provide an artificial intelligence elevator dispatching system to at least partially solve the above technical problems.

(II) technical scheme

To achieve the above object, the present disclosure provides an artificial intelligence elevator dispatching system, comprising:

a plurality of artificial intelligence cameras, configure to and install in the different floors of a building for take the outside wait personnel of elevator, and output statistics number, artificial intelligence camera includes:

the camera shooting piece is used for shooting external images and/or videos;

the processor is used for converting the image and/or the video into a face recognition result, and performing neural network operation by taking the face recognition result as at least part of input data, wherein the output neurons after operation comprise the number of people in the image and/or the video;

an artificial intelligence elevator dispatching device for receiving user request data for at least one floor to determine an elevator dispatching plan in response to a user request to call a floor, the artificial intelligence dispatching device comprising:

the processing chip is used for receiving user request data and carrying out neural network operation by using the user request data, and the output neuron after operation comprises an execution queue of the current user request, wherein the user request data comprises the statistical population data of a user request floor;

and the arithmetic unit determines an elevator dispatching scheme according to the execution queue requested by at least one user.

In a further embodiment, the processor comprises: the storage unit is used for storing the input data, the neural network parameters and the instructions; the control unit is used for reading the special instruction from the storage unit, decoding the special instruction into an arithmetic unit instruction and inputting the arithmetic unit instruction to the arithmetic unit; and the operation unit is used for executing corresponding neural network operation on the data according to the operation unit instruction to obtain an output neuron.

In a further embodiment, in the arithmetic unit, performing the respective neural network operation includes: multiplying the input neuron by the weight data to obtain a multiplication result; executing addition tree operation for adding the multiplication results step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum; and executing activation function operation on the weighted sum which is biased or not processed to obtain the output neuron.

In a further embodiment, the processor comprises: and the preprocessing unit is used for preprocessing the image and/or video data shot by the camera and converting the image and/or video data into a face recognition result, and the face recognition result is data in accordance with a neural network input format.

In a further embodiment, the pre-processing unit includes image and/or video data segmentation, gaussian filtering, binarization, regularization and/or normalization of the captured image and/or video data by the camera to obtain data conforming to a neural network input format.

In a further embodiment, the processor further comprises: and the direct memory access DMA is used for storing the input data, the neural network parameters and the instructions in the storage unit so as to be called by the control unit and the operation unit.

In a further embodiment, the processor further comprises: and the instruction cache is used for accessing the DMA cache instruction from the direct memory for the control unit to call.

In a further embodiment, the neural network parameters include input neurons, output neurons, and weights, and the processor further comprises: the input neuron cache is used for inputting neurons from the direct memory access DMA cache for being called by the operation unit; the weight cache is used for accessing the DMA cache weight from the direct memory for the calling of the arithmetic unit; and the output neuron cache is used for storing the output neurons obtained from the operation unit after operation so as to output the output neurons to the direct memory access DMA.

In a further embodiment, the instruction cache, input neuron cache, weight cache, and output neuron cache are on-chip caches.

In a further embodiment, the system further comprises a transmission unit for transmitting the calculated number of people to the artificial intelligent elevator dispatching equipment in a wireless and/or wired mode.

In a further embodiment, the model for performing the neural network operation in the processing chip is an LSTM neural network model.

In a further embodiment, the neural network operations in the processing chip include: initializing parameters through an LSTM model, obtaining the scheduling cost of a loss function according to user request data, calculating the gradient direction of the minimum cost of adding a user into an execution queue, and outputting the execution queue of a current user request group.

In a further embodiment, the cost of scheduling the penalty function is a weighted average of the user latencies of each floor in the elevator execution queue.

In a further embodiment, the penalty function scheduling cost is the sum of the total number of floors up and down the elevator.

In a further embodiment, the user request data of the processing chip is obtained from a surveillance camera, a mobile phone, a computer, a notebook or a tablet computer.

In a further embodiment, further comprising: and the request signal encoder is used for encoding user request data for the processing chip to call.

In a further embodiment, further comprising: and the memory is used for storing the execution queue of the user request output by the processing chip.

In a further embodiment, determining in the operator an elevator dispatching plan based on the execution queue of at least one user request comprises: and counting the types and the number of the execution queues in the memory, calculating the total busyness of the elevator and the centralized situation of people flow, and determining the total elevator dispatching scheme.

In a further embodiment, the elevator dispatching system further comprises a digital-to-analog converter for converting the digital signal of the elevator dispatching plan into an analog signal for controlling the operation of the elevator motor.

In a further embodiment, further comprising: the input/output unit is used for receiving the signal of the request signal encoder and transmitting the signal into the processing chip as input; the elevator execution queue is also used for receiving the elevator execution queue output by the output end of the processing chip and storing the execution queue into the memory; and the device is also used for reading the original task queue from the memory, inputting the original task queue to the arithmetic unit, and inputting the output result of the arithmetic unit to the digital-to-analog signal converter.

(III) advantageous effects

The artificial intelligent camera in the artificial intelligent elevator dispatching system can analyze the number of people in the shot pictures and/or videos at the camera end, and provides a basis for later analysis;

the artificial intelligence elevator dispatching equipment in the artificial intelligence elevator dispatching system can comprehensively analyze data including waiting number of people and the like in a deep learning mode, and gives a dispatching scheme, so that elevator regulation and control are more accurate and efficient.

Drawings

Fig. 1 is a schematic cross-sectional view of an artificial intelligence camera according to an embodiment of the disclosure.

FIG. 2 is a block diagram of a processor of one embodiment of FIG. 1.

FIG. 3 is a block diagram of a processor of another embodiment of FIG. 1.

Fig. 4 is a diagram of an application scenario of an elevator dispatching system according to an embodiment of the disclosure.

Fig. 5 is a schematic diagram of an elevator dispatching system of an embodiment of the disclosure.

Fig. 6 is a block schematic diagram of an intelligent elevator dispatching device of an embodiment of the present disclosure.

FIG. 7 is a block diagram of a processing chip of one embodiment in FIG. 6.

FIG. 8 is a block diagram of a processing chip of another embodiment of FIG. 6.

Fig. 9 is a schematic diagram of the operation of the neural network of an embodiment of the intelligent elevator dispatching device of fig. 6.

Fig. 10 is a flowchart of the operation of an intelligent elevator dispatching device according to an embodiment of the present disclosure.

Detailed Description

The technical solutions in the embodiments of the present disclosure are clearly and completely described below with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are only a part of the embodiments of the present disclosure, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the disclosure without making any creative effort, shall fall within the protection scope of the disclosure.

A series of solutions of the disclosed embodiments provide a camera capable of performing artificial neural network operations, and an overall elevator dispatching system including the camera. Through the artificial intelligence camera of this disclosed embodiment, can be at the camera head end can the number of people in the picture and/or the video of can the analysis shooting, provide the basis for later stage analysis.

Fig. 1 is a schematic cross-sectional view of an artificial intelligence camera according to an embodiment of the disclosure. As shown in fig. 1, an artificial intelligence camera 100 of the embodiment of the present disclosure includes: a camera 101 and a processor 102, wherein the camera 101 is used for capturing external images and/or videos; the processor 102 is configured to convert the image and/or video into a face recognition result, and perform a neural network operation using the face recognition result as at least part of input data, where an output neuron after the operation includes the number of people in the image and/or video.

The camera 101 may be any of various existing cameras capable of capturing images and/or videos in the prior art, and acquires external information through an electromagnetic or optical or other signal source, and the structure of the camera may refer to various cameras in the prior art, but the camera in the prior art does not include a functional unit or module for counting the number of people after performing neural network operation on image data.

In the embodiment of the present disclosure, the processor 102 is configured to process an image or a video frame captured by the image capturing device 101, and obtain a people counting result in the image or the video frame by performing an artificial neural network operation on a hardware circuit thereof. Preferably, the processor is an artificial neural network chip capable of performing neural network operations.

The image and/or video is converted into a face recognition result, which can be understood as data conforming to a neural network input format, namely input neuron data of an input layer; in the operation of the neural network, the network model used may be various models known in the art, including but not limited to RNN (recurrent neural network) (e.g., LSTM long short term memory network), CNN (convolutional neural network), or DNN (deep neural network), and the demographic data in the image or video frame is contained in the neurons of the output layer of the neural network.

FIG. 2 is a block diagram of a processor of one embodiment of FIG. 1. As shown in fig. 2, in some embodiments, a processor includes a memory unit for storing input data (which may be input neurons), neural network parameters, and instructions; the control unit is used for reading the special instruction from the storage unit, decoding the special instruction into an arithmetic unit instruction and inputting the arithmetic unit instruction to the arithmetic unit; and the arithmetic unit is used for executing corresponding neural network operation on the data according to the arithmetic unit instruction to obtain an output neuron. The storage unit can also store output neurons obtained after operation of the operation unit. Neural network parameters herein include, but are not limited to, weights, biases, and activation functions. Preferably, the initialization weight in the parameter is a trained face recognition weight, and artificial neural network operation (i.e. face recognition calculation (inference)) can be directly performed, so that the process of training the neural network is saved.

In some embodiments, performing the respective neural network operations in the operation unit includes: multiplying the input neuron by the weight data to obtain a multiplication result; executing addition tree operation for adding the multiplication results step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum; and executing activation function operation on the weighted sum which is biased or not processed to obtain the output neuron. Preferably, the activation function may be a sigmoid function, a tanh function, a ReLU function, or a softmax function.

In some embodiments, as shown in fig. 2, the processor may further include a Direct Memory Access (DMA) for storing the input data, the neural network parameters, and the instructions in the storage unit, so as to be called by the control unit and the arithmetic unit; and the operation unit is further used for writing the output neuron into the storage unit after the output neuron is calculated by the operation unit.

In some embodiments, as shown in FIG. 2, the processor further includes an instruction cache for accessing DMA cache instructions from the direct memory for invocation by the control unit. The instruction cache can be an on-chip cache, is integrated on the processor through a preparation process, and can improve the processing speed and save the whole operation time when an instruction is called.

In some embodiments, the processor further comprises: an input neuron cache for inputting neurons from the Direct Memory Access (DMA) cache for invocation by an arithmetic unit; a weight cache for accessing DMA cache weights from the direct memory for the call of an arithmetic unit; an output neuron buffer for storing the output neurons obtained from the operation unit after the operation for output to a Direct Memory Access (DMA). The input neuron cache, the weight cache and the output neuron cache can also be on-chip caches, are integrated on a processor through a semiconductor process, can improve the processing speed when an arithmetic unit reads and writes, and save the whole arithmetic time.

FIG. 3 is a block diagram of a processor of another embodiment of FIG. 1. As shown in fig. 3, the processor in this embodiment may include a preprocessing unit, which is configured to preprocess the image and/or video data captured by the camera and convert the preprocessed image and/or video data into a face recognition result, where the face recognition result is data conforming to a neural network input format. Preferably, the preprocessing comprises segmentation, gaussian filtering, binarization, regularization and/or normalization of the image and/or video data captured by the camera to obtain data conforming to the input format of the neural network. The preprocessing function is to improve the accuracy of the subsequent neural network operation so as to obtain accurate people number judgment.

In some embodiments, the processor may further include a transmission unit for transmitting the calculated number of people data to an external device in a wireless and/or wired manner. The operation is performed by the operation unit after the neural network, and the operation comprises people number judgment data performed on an image or a video frame, and the people number judgment data can be stored in the DMA or the storage unit and can be transmitted to an external device through the transmission unit for further data analysis and application.

The working process of the artificial intelligent camera can be as follows:

and step S1, the camera acquires the elevator opening image signal in real time and converts the elevator opening image signal into an image electric signal.

Step S2, inputting the image electric signal output by the camera to the input end of the processor;

step S3, the image electric signal is processed by the preprocessing unit to form the face recognition result as data conforming to the input format of the neural network, and then the face recognition result is transmitted into the storage unit or the image electric signal is directly transmitted into the storage unit;

step S4, DMA (Direct Memory Access) buffers the instruction, input neurons (including the data conforming to the input format of the neural network) and weight stored in the storage unit, and transmits the buffered instructions, input neurons and weight in batch to the instruction buffer, and inputs the buffered instructions into the neuron buffer and the weight buffer;

step S5, the control unit reads the instruction from the instruction cache, decodes it into the instruction of the arithmetic unit, and then transmits it to the arithmetic unit;

in step S6, according to the arithmetic unit instruction, the arithmetic unit executes the corresponding operation: in each layer of the neural network, the operation can be divided into three steps: s6.1, multiplying the corresponding input neuron by the weight; step S6.2, performing addition tree operation, namely adding the results of the step S6.1 step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum according to needs; and S6.3, performing activation function operation on the result obtained in the step S6.2 to obtain an output neuron, and transmitting the output neuron into an output neuron cache.

And step S7, repeating steps S4-S6, and storing the final output image human face judgment result into the corresponding judgment result storage address in the storage unit by the DMA.

The artificial intelligent camera can be arranged in various occasions (including but not limited to houses, office areas, elevator interiors, elevator exteriors, markets, factories and schools) known in the prior art, is matched with various central control systems for use, and provides real-time data support for people counting for the corresponding systems.

Based on the same inventive concept, the embodiment of the present disclosure further provides an elevator dispatching system, including: the artificial intelligence cameras are configured to be installed on different floors of a building, and used for capturing waiting people outside the elevator and outputting the number of the counted people; and the elevator dispatching equipment responds to a user request for calling a floor, receives input data, counts the number of people of the artificial intelligent camera of the floor called by the input data, and determines an elevator dispatching scheme according to the input data.

Fig. 4 is a diagram of an application scenario of an elevator dispatching system according to an embodiment of the present disclosure, referring to fig. 4, in the building shown in fig. 4, there is at least one elevator 402 (only one elevator is shown in the figure, but obviously, only 1 elevator is shown in the figure), an artificial intelligence camera 401 (there may be multiple cameras on each floor, and preferably, one camera on each floor) described in the above embodiment is installed outside the elevator on each floor, and performs video acquisition on the outside of the elevator, where the acquired video may include an elevator waiting person 403, and for each collected video frame, the intelligent camera 401 on each floor may perform a neural network operation to determine a corresponding number of elevator waiting persons.

Fig. 5 is a schematic diagram of an elevator dispatching system according to an embodiment of the present disclosure, and as shown in fig. 5, the elevator dispatching device 502 and the smart cameras 5011 to 501n (n is a natural number greater than 1) may communicate with each other in a wireless and/or wired manner, for example, the elevator dispatching device 502 sends a control signal to the smart cameras, and the smart cameras send elevator waiting data to the elevator dispatching device 502. Where the corresponding smart camera is labeled, it may indicate where the camera is located, for example, the smart camera 5011 indicates that it is located in the first layer and captures a video of the first layer, and then 501n indicates that the corresponding camera is located in the nth layer and captures a video of the nth layer.

When the elevator dispatching equipment 502 receives an elevator taking request (for example, an elevator waiting person presses an elevator up key or an elevator down key in a corridor), the elevator dispatching system performs calculation according to a set dispatching algorithm, factors considered in the calculation process comprise the number of elevator waiting persons of a floor requested by an elevator, and an elevator dispatching scheme is formed after calculation so as to control the operation of an elevator group motor.

Another group of schemes of the present disclosure will be introduced below, and an artificial intelligence elevator dispatching device (specifically named artificial intelligence elevator dispatching device, which can be distinguished from conventional elevator dispatching devices by internal circuit neural network operation) is provided, which includes a processing device capable of performing artificial neural network operation, and the number of people waiting for an elevator is considered in the artificial neural network operation, so that the formed dispatching scheme is more efficient and accurate. An elevator dispatching system comprising the artificial intelligence elevator dispatching equipment is also provided.

Fig. 6 is a block schematic diagram of an artificial intelligence elevator dispatching device of an embodiment of the disclosure. As shown in fig. 6, the artificial intelligence elevator dispatching device 600 may include a processing chip 601 and an operator 604, where the processing chip 601 is configured to receive a user request data and perform a neural network operation with the user request data, and the output neuron after the operation includes an execution queue of a current user request, where the user request data includes data of the number of elevator waiting persons on a floor requested by a user; and an arithmetic unit 4 for determining an elevator dispatching plan according to the execution queue requested by a plurality of users. Wherein the user request data comprises the number data of elevator waiting persons of the floor requested by the user. In the operation, the operation is performed by the set neural network model, the network model used may be various models existing in the prior art, including but not limited to RNN (recurrent neural network), CNN (convolutional neural network) or DNN (deep neural network), and preferably, the operation is performed by using the LSTM long-short term memory network model in RNN (recurrent neural network).

The user request data includes, but is not limited to, a group of digital codes including one or more requested floors and uplink and downlink, that is, the floors requested by the elevator and the uplink or downlink are digitized by the codes. The user request data may be raw input data or data processed by the intelligent elevator dispatching device 600 according to the embodiment of the disclosure.

The processing chip 601 of the embodiment of the present disclosure may include several functional modules therein, as shown in fig. 7, the processing chip 601 may include a storage module, a control module and an operation module, where the storage module is used to store user request data (which may be used as input neurons), neural network parameters and instructions; the control module is used for reading the special instruction from the storage module, decoding the special instruction into an operation module instruction and inputting the operation module instruction into the operation module; and the operation module is used for executing corresponding neural network operation on the data according to the instruction of the operation module to obtain an output neuron. The storage module can also store output neurons obtained after operation of the operation module. Neural network parameters herein include, but are not limited to, weights, biases, and activation functions. Preferably, the initialization weight in the parameter is a trained face recognition weight or is trained by previous user request data.

In some embodiments, the performing the respective neural network operations in the operation module includes: multiplying the input neuron by the weight data to obtain a multiplication result; executing addition tree operation for adding the multiplication results step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum;

and executing activation function operation on the weighted sum which is biased or not processed to obtain the output neuron. Preferably, the activation function may be a sigmoid function, a tanh function, a ReLU function, or a softmax function.

In some embodiments, as shown in fig. 7, the processing chip 601 may further include a direct memory access module, configured to store the input data, the neural network parameters, and the instructions in the storage module, so as to be called by the control module and the operation module; and the operation module is further used for writing the output neurons into the storage module after the operation module calculates the output neurons.

In some embodiments, as shown in fig. 7, the processing chip 601 further includes an instruction cache module, configured to cache an instruction from the direct memory access module for the control module to call. The instruction cache module can be an on-chip cache, is integrated on the processing chip 601 through a preparation process, and can improve the processing speed and save the whole operation time when an instruction is called.

In some embodiments, the processing chip 601 further comprises: the input neuron cache module is used for caching input neurons from the direct memory access module for being called by the operation module; the weight caching module is used for caching the weight from the direct memory access module for the calling of the operation module; and the output neuron cache module is used for storing the output neurons obtained after the operation by the operation module so as to output the output neurons to the direct memory access module. The input neuron cache module, the weight cache module and the output neuron cache module can also be on-chip caches, and are integrated on the processing chip 601 through a semiconductor process, so that the processing speed can be increased when the operation module reads and writes, and the overall operation time is saved.

Fig. 8 is a block diagram of another embodiment of the processing chip 601 in fig. 6. As shown in fig. 8, the processing chip 601 in this embodiment may include a preprocessing module, which is used to preprocess the user request data and convert the data into data conforming to the input format of the neural network. The preprocessing function is to improve the accuracy of subsequent neural network operation so as to obtain accurate elevator dispatching scheme data.

A typical network model of the processing chip in the embodiment of the present disclosure for performing neural network operations is shown in fig. 9, where the neural network includes an input layer, a plurality of intermediate layers, and an output layer, where circles in the diagram represent neurons, and the neurons of the input layer input user request data (the data includes digital information of one or more user request queues, such as user request queues formed after various encoding), where the user request data includes the number of people waiting for an elevator, and after a plurality of hidden layers of operations, the output layer data includes an execution queue of a current user request group.

Preferably, the neural network model may adopt an LSTM model, where the LSTM model is composed of a plurality of multilayer perceptron models, each multilayer perceptron model represents an LSTM state at a time, the state at the current time is determined by the state at the first n-1 times and the current input, and when the user request data xi is continuously input into the network, the output ht is a user request queue at each time. And moreover, parameters can be initialized through an LSTM model, the scheduling cost of a loss function can be obtained according to relevant user request data, and the execution queue of the current user request group can be output in the gradient direction of the minimum cost for adding users into the execution queue.

Preferably, parameters (such as weight, offset, etc.) in the LSTM (long short term memory network) may be adaptively trained, and then new LSTM model parameters may be generated and trained. Preferably, the adaptive training process is processed in real time.

Preferably, the scheduling cost of the loss function may refer to a weighted average of the waiting time of each floor of the elevator execution queue, that is, a statistic of the average waiting time of all the users requesting the elevator to use for the optimization target (the waiting time is shortest), wherein the assigned weight is the user importance level. The loss function scheduling cost can also be the sum of the total floor numbers of the up-going and down-going elevators, namely the optimization target is the total distance (energy consumption minimum) of elevator operation and the like.

The optimal scheduling cost of the loss function can be a weighting parameter for each elevator request according to the number of people identified by the intelligent camera, so that the command priority is measured by the number of people, and the method is reasonable in practical application.

In some embodiments, the user request data (including the number of people waiting for the elevator) of the processing chip 601 may be from a monitoring camera, a mobile phone, a computer, a notebook, a tablet computer, or other continuous-time image capturing device.

As shown in fig. 6, in some embodiments, the artificial intelligence elevator dispatching device 600 can include a request signal encoder 602 for encoding user request data, such as electrical signal inputs to elevator buttons, into digital information that can be processed by a neural network. For example, the user request in the input user request group is represented by 1 for ascending, and is represented by 0 for descending (of course, the reverse is also possible, 0 for ascending, and 1 for descending), and the input code binary request is composed of the binary code of the floor number and the requested number of people of the elevator at each floor identified by the intelligent camera. Of course, other encoding schemes may be used to form the request code including the request direction and the request floor or other information as input.

As shown in fig. 6, in some embodiments, the artificial intelligence elevator dispatching device 600 may further include a memory 603 for storing an execution queue of user requests output by the processing chip 601, where the execution queue may be multiple (e.g., multiple floors with elevator demands), and further analysis is needed to provide a comprehensive dispatching scheme.

As shown in fig. 6, in some embodiments, the operator 604 is used to count the types and numbers of execution queues in the memory 603, calculate the total busy degree (number of instructions) of the elevator and the concentration of people (stopping condition of each floor), and calculate the total elevator dispatching scheme (e.g., the number of elevators requiring call-up, the specific stopping floor of each elevator, etc.). For example, the operator 604 may perform the following operations: if the number of times of occurrence of each floor in the request queue in a period of time (the time is delta t) before the current time can be counted, and the number is used as the busyness of each floor (the busyness can be normalized into a percentage by multiplying and dividing a constant term); and meanwhile, counting the total request number in a period of time (the time can be delta t) before the current moment, wherein the total request number is used as the total busy degree of the elevator (the total busy degree can be normalized into a percentage through a multiplication and division constant term). The total busy degree of the elevator determines the working number of the elevator boxes, such as 0-25% of the total busy degree; 1 ladder box runs, 25% -50%, two ladder boxes runs, 50% -75%, three, 75% -100%, and four. Wherein the threshold value of 25%, 50% or 75% can be adjusted according to actual conditions. The busyness of each elevator layer is taken as the weight of the elevator dispatching loss function of the elevator layer, so that the elevator dispatching loss function is more busyness in visual understanding, the weight is larger, the influence in the loss function is larger, and the loss function can be considered preferentially when being optimized.

As shown in fig. 6, in some embodiments, the artificial intelligence elevator dispatching device 600 can also include a digital-to-analog converter 606 that converts digital signals representing an elevator dispatching plan into analog signals that control elevator motor operation.

As shown in fig. 6, in some embodiments, the artificial intelligence elevator dispatching device 600 can include an input/output unit 605, the input/output unit 605 for receiving the signal of the request signal encoder 602 and passing it into the processing chip 601 as an input; and is also used for receiving the output of the output end of the processing chip (outputting an elevator execution queue), and storing the execution queue into the memory 603; and the system is also used for reading an original task queue from the memory 603, inputting the original task queue into the arithmetic unit 604, counting the busyness and the people flow concentration of the elevator, and inputting an output result into the digital-to-analog signal converter 606 to control the running number and the parking floor of the elevator so as to intelligently control the running of the elevator in real time.

As shown in fig. 10, the operation method of the above-described artificial intelligence elevator dispatching device 600 may include:

step S101, user request data is transmitted into a storage module through a preprocessing module or directly transmitted into the storage module;

step S102, the direct memory access module transmits the instruction to the instruction cache module in batches, and inputs the instruction to the neuron cache module and the weight cache module;

step S103, the control module reads the instruction from the instruction cache module, decodes the instruction and transmits the decoded instruction to the operation module;

step S104, according to the instruction, the operation module executes the corresponding operation: in each layer of the neural network, the operation is mainly divided into three steps: step S104.1, multiplying the corresponding input neuron by the weight; step S104.2, performing addition tree operation, namely adding the results of the step S104.1 step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum according to needs; and step S104.3, performing activating function operation on the result obtained in the step S104.2 to obtain an output neuron, and transmitting the output neuron into an output neuron cache.

Step S105, repeating the steps S102 to S104 until all data are operated;

and step S106, storing the result after the operation as an elevator execution ordered sequence into a corresponding judgment result storage address by the direct memory access module, and outputting the judgment result storage address to the storage 603.

And S107, temporarily storing and counting the busyness and the concentrated situation of the people flow of the elevator at the time interval through an elevator control queue of the artificial intelligent elevator dispatching equipment 600, determining the number of running elevators according to the threshold classification of the busyness, determining the stopped or started elevator according to the concentrated situation of the people flow, and then sending an elevator control electric signal to control the running of the elevator.

The other series of the schemes of the embodiment of the disclosure provide an artificial intelligence elevator dispatching system, which comprises a plurality of artificial intelligence cameras of the embodiment and the artificial intelligence elevator dispatching equipment of the embodiment, and the accuracy and the timeliness of elevator dispatching can be integrally improved by identifying the number of people through artificial intelligence and analyzing the dispatching scheme through the artificial intelligence.

The artificial intelligence camera of the embodiment of the disclosure is internally provided with a processor capable of operating through a neural network, the number of people waiting for taking an elevator at an elevator entrance can be identified in real time according to the camera shooting through the neural network, finally, the number of people requesting floors, the number of ascending and descending marks and the number of people waiting for taking the elevator are input into artificial intelligence elevator dispatching equipment of an elevator dispatching system as output to be coded, the input is converted into a recognizable signal of a processing chip and transmitted into the processing chip to construct an elevator dispatching instruction queue, the processed instruction queues are transmitted into the artificial intelligence elevator dispatching equipment to be decoded, and the recognizable signal is converted into an analog electric signal through a digital-to-analog converter to directly control.

The artificial intelligence camera of the disclosed embodiments can be configured and implement corresponding functions with reference to the embodiments described above in connection with fig. 1-4; the artificial intelligence elevator dispatching equipment of the embodiment of the disclosure can be set and realize corresponding functions by referring to the artificial intelligence elevator dispatching equipment combined with the devices shown in fig. 6-10, and details are not repeated herein.

In the embodiments provided in the present disclosure, it should be understood that the disclosed related devices and methods may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the described parts or modules is only one logical division, and other divisions may be realized in practice, for example, a plurality of parts or modules may be combined or integrated into a system, or some features may be omitted or not executed.

In this disclosure, the term "and/or" may have been used. As used herein, the term "and/or" means one or the other or both (e.g., a and/or B means a or B or both a and B).

In the description above, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the disclosure. It will be apparent, however, to one skilled in the art, that one or more other embodiments may be practiced without some of these specific details. The specific embodiments described are not intended to limit the disclosure but rather to illustrate it. The scope of the present disclosure is not to be determined by the specific examples provided above but only by the claims below. In other instances, well-known circuits, structures, devices, and operations are shown in block diagram form, rather than in detail, in order not to obscure an understanding of the description. Where considered appropriate, reference numerals or terminal portions of reference numerals have been repeated among the figures to indicate corresponding or analogous elements, optionally having similar characteristics or identical features, unless otherwise specified or evident.

Various operations and methods have been described. Some methods have been described in a relatively basic manner in a flow chart form, but operations may alternatively be added to and/or removed from the methods. Additionally, while the flow diagrams illustrate a particular order of operation according to example embodiments, it is understood that this particular order is exemplary. Alternative embodiments may optionally perform these operations in a different manner, combine certain operations, interleave certain operations, etc. The components, features, and specific optional details of the devices described herein may also optionally be applied to the methods described herein, which may be performed by and/or within such devices in various embodiments.

Each functional unit/subunit/module/submodule in the present disclosure may be hardware, for example, the hardware may be a circuit, including a digital circuit, an analog circuit, and the like. Physical implementations of hardware structures include, but are not limited to, physical devices including, but not limited to, transistors, memristors, and the like. The computing module in the computing device may be any suitable hardware processor, such as a CPU, GPU, FPGA, DSP, ASIC, and the like. The memory unit may be any suitable magnetic or magneto-optical storage medium, such as RRAM, DRAM, SRAM, EDRAM, HBM, HMC, etc.

It will be clear to those skilled in the art that, for convenience and simplicity of description, the foregoing division of the functional modules is merely used as an example, and in practical applications, the above function distribution may be performed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to perform all or part of the above described functions.

The above-mentioned embodiments are intended to illustrate the objects, aspects and advantages of the present disclosure in further detail, and it should be understood that the above-mentioned embodiments are only illustrative of the present disclosure and are not intended to limit the present disclosure, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.

Claims

1. An artificial intelligence elevator dispatching system, characterized by comprising:

the camera shooting piece is used for shooting external images and/or videos;

an artificial intelligence elevator dispatching device for receiving user request data for at least one floor to determine an elevator dispatching plan in response to a user request to call a floor, the artificial intelligence elevator dispatching device comprising:

the processing chip is used for receiving user request data and carrying out neural network operation by using the user request data, and the output neuron after operation comprises an execution queue of the current user request, wherein the user request data comprises the statistical population data of a user request floor; wherein, the model for carrying out the neural network operation in the processing chip is an LSTM neural network model, and the neural network operation in the processing chip comprises the following steps:

initializing parameters through an LSTM model, obtaining the scheduling cost of a loss function according to user request data, calculating the gradient direction of the minimum cost of adding a user into an execution queue, and outputting the execution queue of a current user request group;

2. The artificial intelligence elevator dispatching system of claim 1, wherein the processor comprises:

the storage unit is used for storing the input data, the neural network parameters and the instructions;

the control unit is used for reading the special instruction from the storage unit, decoding the special instruction into an arithmetic unit instruction and inputting the arithmetic unit instruction to the arithmetic unit;

and the operation unit is used for executing corresponding neural network operation on the data according to the operation unit instruction to obtain an output neuron.

3. The artificial intelligence elevator dispatching system of claim 2, wherein the arithmetic unit wherein performing the respective neural network operation comprises:

multiplying the input neuron by the weight data to obtain a multiplication result;

executing addition tree operation for adding the multiplication results step by step through an addition tree to obtain a weighted sum, and adding bias or not processing the weighted sum;

and executing activation function operation on the weighted sum which is biased or not processed to obtain the output neuron.

4. The artificial intelligence elevator dispatching system of claim 1, wherein the processor comprises:

and the preprocessing unit is used for preprocessing the image and/or video data shot by the camera and converting the image and/or video data into a face recognition result, and the face recognition result is data in accordance with a neural network input format.

5. The artificial intelligence elevator dispatching system of claim 4, wherein in the preprocessing unit, the preprocessing comprises image and/or video data segmentation, Gaussian filtering, binarization, regularization and/or normalization of the images captured by the cameras to obtain data that conforms to a neural network input format.

6. The artificial intelligence elevator dispatching system of claim 2, wherein the processor further comprises:

and the direct memory access DMA is used for storing the input data, the neural network parameters and the instructions in the storage unit so as to be called by the control unit and the operation unit.

7. The artificial intelligence elevator dispatching system of claim 6, wherein the processor further comprises:

and the instruction cache is used for accessing the DMA cache instruction from the direct memory for the control unit to call.

8. The artificial intelligence elevator dispatching system of claim 7, wherein the neural network parameters include input neurons, output neurons, and weights, the processor further comprising:

the input neuron cache is used for inputting neurons from the direct memory access DMA cache for being called by the operation unit;

the weight cache is used for accessing the DMA cache weight from the direct memory for the calling of the arithmetic unit;

and the output neuron cache is used for storing the output neurons obtained from the operation unit after operation so as to output the output neurons to the direct memory access DMA.

9. The artificial intelligence elevator dispatching system of claim 8, wherein the instruction cache, input neuron cache, weight cache, and output neuron cache are on-chip caches.

10. The artificial intelligence elevator dispatching system of claim 1, further comprising a transmission unit for transmitting the calculated number of people data to an artificial intelligence elevator dispatching device in a wireless and/or wired manner.

11. The artificial intelligence elevator dispatching system of claim 1, wherein the dispatching cost of the penalty function is a weighted average of user wait times at each floor in an elevator execution queue.

12. The artificial intelligence elevator dispatching system of claim 1, wherein the penalty function dispatching cost is a sum of total floor numbers of up and down elevators.

13. The artificial intelligence elevator dispatching system of claim 1, wherein the user request data of the processing chip is obtained from a surveillance camera, a cell phone, a computer, a notebook, or a tablet computer.

14. The artificial intelligence elevator dispatching system of claim 1, further comprising:

and the request signal encoder is used for encoding user request data for the processing chip to call.

15. The artificial intelligence elevator dispatching system of claim 14, further comprising:

and the memory is used for storing the execution queue of the user request output by the processing chip.

16. The artificial intelligence elevator dispatching system of claim 15, wherein determining an elevator dispatching plan in the operator based on the execution queue of at least one user request comprises:

and counting the types and the number of the execution queues in the memory, calculating the total busyness of the elevator and the centralized situation of people flow, and determining the total elevator dispatching scheme.

17. The artificial intelligence elevator dispatching system of claim 16, further comprising a digital-to-analog converter for converting digital signals of the elevator dispatching plan to analog signals for controlling operation of an elevator motor.

18. The artificial intelligence elevator dispatching system of claim 17, further comprising:

the input/output unit is used for receiving the signal of the request signal encoder and transmitting the signal into the processing chip as input; the elevator execution queue is also used for receiving the elevator execution queue output by the output end of the processing chip and storing the execution queue into the memory; and the device is also used for reading the original task queue from the memory, inputting the original task queue to the arithmetic unit, and inputting the output result of the arithmetic unit to the digital-to-analog signal converter.