Intelligent store management method and system based on big data

Abstract

The invention belongs to the technical field of intelligent monitoring, and provides an intelligent store management method and system based on big data, and the method comprises the steps: obtaining a monitoring video outside a store through a camera, obtaining the behavior information of each moving target in the video, carrying out the recognition and analysis of the behavior characteristics of the moving targets according to the behavior information, and carrying out the recognition and analysis of the behavior characteristics. The working state of the store is intelligently controlled according to the recognition and analysis result, the intelligent management effect of the store can be remarkably improved, meanwhile, the behavior of the target outside the store can be accurately predicted by recognizing and analyzing the behavior characteristics of the moving target, and the defect that current intelligent store management only aims at the oneness in the store is overcome.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent monitoring technology, and specifically relates to an intelligent store management method and system based on big data. Background Art

[0002] Smart stores are a retail model that uses advanced technology and digital means to improve the efficiency of retail operations, optimize user experience, and enhance market competitiveness. Online and offline integrated operations and unmanned sales are all operating models of smart stores. For such smart stores, in order to facilitate customers to enter the store, a sensing system is generally installed outside the store. The sensing system itself is equipped with a sensing detector that can emit an infrared or microwave signal. When this signal is reflected by an approaching object, an echo signal will be received at the detector receiving end. According to the echo signal, the closure of the sensing door is controlled to achieve automatic opening and closing. However, the automatic sensing doors on the market are not smart enough. When the sensing detector detects that a moving target is approaching and reaches a preset distance, the automatic sensing door will be opened, and the timing of opening the door cannot be controlled according to different situations. In addition, the current design of the system for multiple people to shop without feeling at the same time usually tracks the items in the store and analyzes the behavior of customers to improve the safety of unmanned supermarkets and reduce the operating risks of unmanned supermarkets. Therefore, there is a lack of recognition and analysis of the behavior of customers outside the store, which will lead to the inability to accurately grasp the situation of customers outside the unmanned supermarket and the inability to predict the behavior of customers in advance. Summary of the invention

[0003] The purpose of the present invention is to propose an intelligent store management method and system based on big data to solve one or more technical problems existing in the prior art and at least provide a beneficial choice or create conditions.

[0004] In order to achieve the above object, according to one aspect of the present invention, a smart store management method based on big data is provided, and the smart store management method based on big data comprises the following steps: S100, obtaining behavior information of a moving target in a surveillance video outside a store; S200, calculating the fuzzy coefficient of the moving target based on the behavior information; S300, marking fuzzy information in the behavior information according to the fuzzy coefficient; S400, calculating the store entry intention of the mobile target according to the fuzzy information; S500, performing a correction count of the flow of people entering the store based on the store entry intention; Furthermore, the store is an unmanned store.

[0005] Furthermore, in S100, a high-definition camera is used to capture a video of the monitored area.

[0006] Furthermore, the high-definition camera adopts a binocular camera.

[0007] Furthermore, the monitoring area is a square or supermarket around the monitoring store.

[0008] At present, the access control management system of smart stores adopts the smart home system, which detects whether there are objects approaching within the range by installing infrared or microwave sensors on the door. However, the above system is prone to misidentification, that is, as long as there is an object moving within the detection range, the sensor door will be opened, especially in hot weather, the sensor door is frequently opened, resulting in air leakage, causing unnecessary losses to the store; and when the camera captures the target, it not only shows the real-time image of a single time, but more often shows the scene over a period of time. When the objects in the scene move, the image of the scene must show the complete combination of all positions of those objects within the exposure time (depending on the shutter speed) and the camera's perspective. In such a picture, any object moving in the direction relative to the camera will appear blurred or shaken. Researchers call this phenomenon motion blur.

[0009] At present, in order to eliminate the problem of dynamic blur, the moving parameters of the camera and the captured target (moving parameters include moving speed and moving direction) are usually kept consistent to eliminate the relative movement error between the camera and the captured target to eliminate the dynamic blur phenomenon in the video. However, when facing the capture of multiple targets, there will be a short-term parameter inconsistency phenomenon, resulting in dynamic blur phenomenon in all objects in a short period of time, which introduces a lot of interference to the subsequent behavior recognition of a single target; or the pixel value where the motion blur is generated is extracted to correct the surrounding pixel values ​​to correct the part of the image where the motion blur is generated. However, the above methods cannot be applied to target capture in complex background environments. To solve the above problems, the present invention provides the following method to quickly eliminate dynamic blur interference in complex environments by calculating the intention to enter the store: Further, in S100, the method for obtaining the behavior information of the mobile target in the surveillance video outside the store is: a plurality of mobile targets are marked in the video captured by the high-definition camera installed outside the store to form a monitoring target set MOT, and i is used as the sequence number of the mobile target, mot i is the i-th moving target; The video processing software Tracker is used to extract the speed and trajectory of the moving target, and the speed of all moving targets is formed into a speed set SPD, spd i is the speed of the ith moving target, calculate the distance between the moving target and the geometric center of the area occupied by the store, and the distances between all moving targets and the geometric center of the area occupied by the store form a distance set DST, dst iis the distance between the i-th moving target and the geometric center of the area occupied by the store. The speed set SPD and the distance set DST together constitute the behavior information of the moving target.

[0010] Furthermore, any one of the frame difference method, background subtraction method, YOLO algorithm, and SSD algorithm is used to mark the moving target.

[0011] Further, in S200, the method for calculating the fuzzy coefficient of the moving target based on the behavior information is: In the speed set, the speed of all moving targets is decomposed. The specific method is: The direction from the target to the geometric center of the area where the store is located is the horizontal direction, and the direction perpendicular to the geometric center of the area where the store is located is the vertical direction. The projection of the speed in the horizontal direction is spd_xi, and the projection of the speed in the vertical direction is spd_yi; Set the projection coefficient to m, and set the initial value of m to 1, calculate the difference between spd_xi and spd_yi, and modify the projection coefficient m according to the difference result; Furthermore, the specific method of modifying the projection coefficient m according to the difference result is: If the difference between spd_xi and spd_yi is less than or equal to zero, then change the value of m to 0, otherwise, do not change the value of m; The video is divided into multiple video images according to the preset time t, and the distance between the moving target and the geometric center of the area occupied by the store is obtained at the same time. The distance of all moving targets and the corresponding acquisition time are plotted to generate a distance change curve XL, where the distance change curve XL is a curve composed of time and distance. The point on the XL curve is called the distance point dip, and k is the serial number of the acquisition time. dip k For T k The distance between the first target obtained at the moment and the geometric center of the area occupied by the store, where T k Indicates the kth acquisition time; Set the distance coefficient to x, and set the initial value of x to 1, calculate the sum of the derivatives at all distance points on the curve XL, and modify the distance coefficient x according to the calculation result; Furthermore, the specific method for modifying the distance coefficient x according to the calculation result is: If the sum of the derivatives at all distance points on the curve XL is less than zero, the value of x is modified to 0, otherwise the value of x is not modified; The speed of the moving target is obtained in the same way. The speed of all moving targets is plotted against the acquisition time to generate a speed change curve VL. The point on the curve VL is called a speed point rat. Let k be the serial number of the acquisition time, rat k For T kThe speed of the moving target obtained at all times; Set the speed coefficient to v, and set the initial value of v to 1, calculate the sum of the derivatives at all speed points on the curve VL, and modify the speed coefficient v according to the calculation result; Furthermore, the specific method of modifying the speed coefficient v according to the calculation result is: If the sum of the derivatives at all speed points on the curve VL is less than zero, the v value is changed to 0, otherwise the v value is not changed; The fuzzy coefficient of the moving target is calculated according to the formula m+x+v.

[0012] Furthermore, in S300, the specific method of marking fuzzy information in behavior information according to the fuzzy coefficient is: all moving targets with fuzzy coefficient equal to zero are recorded as relevant targets, all video images containing relevant targets constitute a behavior set, and the behavior information of the relevant targets is recorded as fuzzy information.

[0013] Furthermore, in S400, the specific method of calculating the mobile target's store-entry intention according to the deblurred information is as follows: Record the speed rat and distance dip of the relevant target collected at the latest time on curves VL and XL, and convert the speed and distance according to the formula P=Ln( ) calculates the store entry intention of the current relevant target, where the Ln() function is a logarithmic function with the natural constant e as the base, where a represents the speed parameter of the relevant target in the acquired video, according to the formula a= , where rum represents the number of images with a speed greater than the average speed in the behavior set, Nnum represents the total number of images in the behavior set; b represents the distance parameter of the relevant target in the acquired video, according to the formula b= , where dum represents the number of images in the behavior set that are less than the average distance.

[0014] The current causes of motion blur include: (1) when the shutter speed is insufficient, the target moves during the exposure period, resulting in pixel displacement; (2) multiple targets move at different speeds and directions, complicating the blur pattern (such as crowds and traffic monitoring scenes). Traditional deblurring algorithms are difficult to confirm because the blur kernels of multiple blurred targets are difficult to confirm. The existing method uses an algorithm to mark the targets and then estimates the blur kernel of each target. However, it is impossible to complete the deblurring of multiple targets in a complex environment, resulting in poor quality of the captured target trajectory. The above method uses the Ln logarithmic function to convert the movement parameters of a single target into a function based on the scaling principle of the function. It can quickly eliminate the movement of a single target during the exposure period when capturing the target. At the same time, the linear fitting method eliminates the phenomenon of temporary disappearance when the camera obtains the movement parameters when multiple moving targets have partial repetition in the path. It can quickly correct the target's movement trajectory in a short time without the need for an anti-shake hardware module, significantly improving the practicality of real-time analysis of the target.

[0015] However, the above method only draws the trajectory based on the speed and distance as parameters, and does not take into account the influence of the interaction force on the trajectory, whether it is a single individual or multiple groups, or between individuals and groups, that is, the group attraction effect is generated, and the attraction of different groups to individuals is greater than the attraction of a single group multiple times, that is, individuals will produce a replication effect in multiple groups, see reference: Cluster coding and replication effect of group attraction; however, the above method obviously fails to take into account the group attraction effect and replication effect. In practical applications, if the group attraction effect and replication effect are not taken into account, the captured trajectory will show a missing phenomenon, that is, the trajectory obtained at this time has a slight difference from the actual trajectory. In order to eliminate the interference of the group attraction effect and replication effect on behavioral information, the present invention proposes the following preferred solution to calculate the store entry intention: In the video, the crowd density calculation algorithm is used to mark multiple groups of people to form a set GP, and j is used as the sequence number of the group. j Indicates the jth group of people. In the monitoring target set MOT, the curve ATY formed by the action trajectory of each moving target is obtained in turn, where the curve ATY is a curve composed of the modulus of the time and the velocity vector. The velocity vector of the moving target is obtained as the target vector according to the preset period t, and the velocity vectors of all people are obtained as the influence vector to form an influence set; Furthermore, the crowd density calculation algorithm is any one of the models of MCNN, CrowdNet, Slicing CNN, and ConvLSTM.

[0016] Traverse all velocity vectors on the curve ATY. If the velocity vector modulus at the current moment is smaller than the velocity vector modulus at the previous moment and also smaller than the velocity vector modulus at the next moment, or if the velocity vector modulus at the current moment is larger than the velocity vector modulus at the previous moment and also larger than the velocity vector modulus at the next moment, then mark the velocity vector at this moment as an affected vector. In the impact set, subtract all affected vectors from the impact vector in turn to find the magnitude of the vector modulus. Record the crowd corresponding to the impact vector with the largest modulus as the main impact group. Calculate the distance between the geometric center of the main impact group and the current moving target as the reference moment at the current time point. Calculate the distance between the geometric center of all crowds and the moving target and the magnitude of the reference moment. If it is larger than the reference moment, then record the current crowd as a weak impact group, otherwise, record it as a strong impact group. Traverse all the marked crowds in turn. If it is a strong influence group, subtract the velocity vector of the moving target at the current moment from the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. If it is a weak influence group, add the target vector of the moving target at the current moment to the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. The ratio of rat_h to the distance between the moving target and the geometric center of the area occupied by the store (the numerator of the calculated ratio is rat_h, and the denominator of the calculated ratio is the distance between the moving target and the geometric center of the area occupied by the store) is used as the local store-entering intention of the crowd for the moving target at the current moment. According to the above method, the local store-entering intention of the crowd for the moving target at all moments is calculated, and the sum of all local store-entering intentions is calculated as the store-entering intention of the current moving target.

[0017] The above method detects and updates the main influence group in real time, divides the remaining people into strong influence groups and weak influence groups according to the main influence group, and then calculates the influence of all people on individuals in real time, thereby avoiding the replication effect of multiple groups on individuals. By performing feature recognition on the behavioral information detected at this time, it is possible to more accurately predict the behavior in a more complex environment. At the same time, the above method corrects the mutual influence between the detection targets in a complex environment, thereby eliminating the errors in the parameters, and also repairs the slight differences in the movement trajectory of the target drawn according to the parameters, thereby realizing accurate and rapid analysis and recognition of the target's characteristic behavior.

[0018] Although the above methods can achieve real-time prediction of moving targets, they still cannot solve the problem of being pinched by the induction door due to inconsistent target entry time. At present, the problem of being pinched by the induction door is generally solved by continuously sensing whether there is a biological signal between the doors through the infrared sensor on the induction door. If there is, the induction door is kept open. If not, the induction door is controlled to be closed after a period of time. However, this method cannot guarantee good safety and real-time performance. If a biological signal is detected all the time, the door will remain open. If the flow of people cannot be controlled during peak hours, trampling or other serious consequences may occur. In order to solve the above problems, real-time counting of the induction door and monitoring of the store are currently used. A method combining human flow statistics with manual order maintenance is used to control the flow of people. However, the sensor door requires the sensor to have extremely high sensitivity to achieve real-time counting. However, the overly sensitive sensor will have glitches and missing pulse waveforms when encountering a swinging object (a swinging object refers to the same object repeatedly sensed by the sensor), resulting in delayed processing or erroneous processing results when the sensor receives the above abnormal pulses, which are extremely likely to interfere with the counting results, especially during peak hours of human flow, which greatly increases the safety risks. For this reason, the present invention solves the above problems by marking the abnormal pulse group in the pulse signal generated by the sensor to assist in monitoring and correcting the abnormal waveform that produces glitches and missing pulses: Furthermore, in S500, the specific method for correcting the counting of the flow of people entering the store based on the intention of entering the store is: The pulse signal generated by the sensor is obtained at the time interval T within the preset period, and the generation time of the effective pulse (effective pulse refers to the pulse signal generated when the sensor receives the echo signal) is recorded as TIM_1, and the corresponding effective pulse disappearance time is TIM_2. A variable M is set, and the initial value of M is set to 0. Whenever a complete set of TIM_1 and TIM_2 is detected, or whenever a complete set of pulse signals is detected, the variable M is increased by 1. After the preset period ends, the value of the variable M is accessed as the store traffic during this period; The acquired pulse signal is plotted as a square wave signal SL, where the square wave signal SL is a curve composed of time and pulse signal. The size of the pulse signal is compared on the curve SL starting from the second acquisition time. If the current signal size is greater than the signal size at the previous moment, the time point corresponding to the current signal is recorded as ST. When searching backward from ST on the curve SL for the first point that satisfies the current signal being greater than the signal at the next moment, the time point corresponding to the searched signal is recorded as ET. The above operation is repeated until all STs and ETs are marked. Starting from the first ST, each ST and the adjacent ET are respectively formed into a valid pulse group. The time interval between ST and ET of each valid pulse group is calculated in turn as the signal duration TIM of the group. With n as the serial number, TIM nIndicates the signal duration of the nth valid pulse group; Traverse all TIMs within the range of n. When TIM satisfies the formula Ln(TIM n )<ε, mark the valid pulse group corresponding to the current TIM as an abnormal pulse group, where ε is an arbitrary positive number, and traverse all abnormal pulse groups. If the formula Ln(TIM n )+ Ln(TIM n+1 )<ε or satisfy the formula Ln(TIM n )+ Ln(TIM n-1 )<ε, the current abnormal pulse group is marked as a repeated pulse group, otherwise the current abnormal pulse group is marked as a lost pulse group; The statistical flow of people entering the store is corrected according to the abnormal pulse group. The specific method is: the flow of people entering the store plus the number of lost pulse groups minus the number of repeated pulse groups is used as the corrected flow of people entering the store.

[0019] The abnormal pulse group in the above method represents the abnormal square wave signal generated by the sensor due to the induction of the swing object. If the current pulse group signal satisfies Ln(TIM n )<ε, it means that multiple objects are sensed at the same time and thus mistakenly identified as one object, resulting in the burr part on the pulse signal. If the current pulse group signal satisfies Ln(TIM n )+ Ln(TIM n+1 )<ε or Ln(TIM n )+ Ln(TIM n-1 )<ε means that the same object is sensed multiple times and the missing part is generated in one pulse signal. By using ε as the marking rule, the response time of the sensor when facing abnormal pulse signals can be eliminated, avoiding the limitations of the above method due to inconsistent sensor performance. The practicality of the method can be significantly improved, and trampling and other accidents caused by excessive human flow can be prevented in time.

[0020] The present invention also provides a smart store management system based on big data, the system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program running in the following system units: A real-time video recording unit is used to record surveillance videos outside the store through cameras outside the store; A behavior information acquisition unit, used to acquire behavior information of a moving target in the captured surveillance video outside the store; A fuzzy coefficient calculation unit, used for calculating the fuzzy coefficient of the moving target according to the acquired behavior information; A fuzzy information marking unit, used for marking the fuzzy information in the behavior information according to the calculated fuzzy coefficient; A store entry intention calculation unit, used to calculate the store entry intention of the mobile target according to the fuzzy information of the mobile target; The store flow counting unit is used to correct the current store flow according to the store entry intention of the mobile target.

[0021] The beneficial effects of the present invention are as follows: the present invention provides an intelligent store management method and system based on big data, which obtains the behavior information of mobile targets outside the store and performs behavioral feature analysis to intelligently control the store, significantly improving the intelligence of store management, while eliminating the interference of group attraction effect and replication effect on behavioral information, and can perform real-time and accurate feature analysis of mobile targets, thereby making up for the singleness of the current intelligent management system that only focuses on in-store management, monitoring the flow of people in the store in real time, and avoiding the hidden danger of trampling during peak hours due to the statistical flow of people not being consistent with the actual flow due to inconsistent sensor performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Shown is a flow chart of a smart store management method based on big data; Figure 2 Shown is a structural diagram of an intelligent store management system based on big data. DETAILED DESCRIPTION

[0023] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0024] Embodiment 1: Figure 1 Shown is a flow chart of an intelligent store management method based on big data.

[0025] Reference Figure 1 The present invention proposes a smart store management method based on big data, which comprises the following steps: S100, obtaining behavior information of a moving target in a surveillance video outside a store; S200, calculating the fuzzy coefficient of the moving target based on the behavior information; S300, marking fuzzy information in the behavior information according to the fuzzy coefficient; S400, calculating the store entry intention of the mobile target according to the fuzzy information; S500, performing a correction count of the flow of people entering the store based on the store entry intention; Furthermore, the store is an unmanned store.

[0026] Furthermore, in S100, a high-definition camera is used to capture a video of the monitored area.

[0027] Furthermore, the high-definition camera adopts a binocular camera.

[0028] Furthermore, the monitoring area is a square or supermarket around the monitoring store.

[0029] Further, in S100, the method for obtaining the behavior information of the mobile target in the surveillance video outside the store is: a plurality of mobile targets are marked in the video captured by the high-definition camera installed outside the store to form a monitoring target set MOT, and i is used as the sequence number of the mobile target, mot i is the i-th moving target; The video processing software Tracker is used to extract the speed and trajectory of the moving target, and the speed of all moving targets is formed into a speed set SPD, spd i is the speed of the ith moving target, calculate the distance between the moving target and the geometric center of the area occupied by the store, and the distances between all moving targets and the geometric center of the area occupied by the store form a distance set DST, dst i is the distance between the i-th moving target and the geometric center of the area occupied by the store. The speed set SPD and the distance set DST together constitute the behavior information of the moving target.

[0030] Furthermore, the YOLO algorithm is used to mark the moving targets.

[0031] Further, in S200, the method for calculating the fuzzy coefficient of the moving target based on the behavior information is: In the speed set, the speed of all moving targets is decomposed. The specific method is: The direction from the target to the geometric center of the area where the store is located is the horizontal direction, and the direction perpendicular to the geometric center of the area where the store is located is the vertical direction. The projection of the speed in the horizontal direction is spd_xi, and the projection of the speed in the vertical direction is spd_yi; Set the projection coefficient to m, and set the initial value of m to 1, calculate the difference between spd_xi and spd_yi, and modify the projection coefficient m according to the difference result; Furthermore, the specific method of modifying the projection coefficient m according to the difference result is: If the difference between spd_xi and spd_yi is less than or equal to zero, then change the value of m to 0, otherwise, do not change the value of m; The video is divided into multiple video images with a period of 2000ms, and the distance between the moving target and the geometric center of the store area is obtained at the same time. The distance of all moving targets and the corresponding acquisition time are plotted to generate a distance change curve XL, where the distance change curve XL is a curve composed of time and distance. The point on the XL curve is called the distance point dip, and k is the serial number of the acquisition time. k For T k The distance between the first target obtained at the moment and the geometric center of the area occupied by the store, where T k Indicates the kth acquisition time; Set the distance coefficient to x, and set the initial value of x to 1, calculate the sum of the derivatives at all distance points on the curve XL, and modify the distance coefficient x according to the calculation result; Furthermore, the specific method for modifying the distance coefficient x according to the calculation result is: If the sum of the derivatives at all distance points on the curve XL is less than zero, the value of x is modified to 0, otherwise the value of x is not modified; The speed of the moving target is obtained in the same way. The speed of all moving targets is plotted against the acquisition time to generate a speed change curve VL. The point on the curve VL is called a speed point rat. Let k be the serial number of the acquisition time, rat k For T k The speed of the moving target obtained at all times; Set the speed coefficient to v, and set the initial value of v to 1, calculate the sum of the derivatives at all speed points on the curve VL, and modify the speed coefficient v according to the calculation result; Furthermore, the specific method of modifying the speed coefficient v according to the calculation result is: If the sum of the derivatives at all speed points on the curve VL is less than zero, the v value is changed to 0, otherwise the v value is not changed; The fuzzy coefficient of the moving target is calculated according to the formula m+x+v.

[0032] Furthermore, in S300, the specific method of marking fuzzy information in behavior information according to the fuzzy coefficient is: all moving targets with fuzzy coefficient equal to zero are recorded as relevant targets, all video images containing relevant targets constitute a behavior set, and the behavior information of the relevant targets is recorded as fuzzy information.

[0033] Furthermore, in S400, the specific method of calculating the mobile target's store-entry intention according to the deblurred information is as follows: Record the speed rat and distance dip of the relevant target collected at the latest time on curves VL and XL, and convert the speed and distance according to the formula P=Ln( ) calculates the store entry intention of the current relevant target, where the Ln() function is a logarithmic function with the natural constant e as the base, where a represents the speed parameter of the relevant target in the acquired video, according to the formula a= , where rum represents the number of images with a speed greater than the average speed in the behavior set, Nnum represents the total number of images in the behavior set; b represents the distance parameter of the relevant target in the acquired video, according to the formula b= , where dum represents the number of images in the behavior set that are less than the average distance.

[0034] Furthermore, in S500, the specific method for correcting the counting of the flow of people entering the store based on the intention of entering the store is: The pulse signal generated by the sensor is obtained periodically every 1 second within every 60 seconds. The generation time of the effective pulse (effective pulse refers to the pulse signal generated when the sensor receives the echo signal) is recorded as TIM_1, and the corresponding effective pulse disappearance time is TIM_2. A variable M is set, and the initial value of M is set to 0. Whenever a complete set of TIM_1 and TIM_2 is detected, or whenever a complete set of pulse signals is detected, the variable M is increased by 1. After the preset period ends, the value of the variable M is accessed as the store traffic during this period; The acquired pulse signal is plotted as a square wave signal SL, where the square wave signal SL is a curve composed of time and pulse signal. The size of the pulse signal is compared on the curve SL starting from the second acquisition time. If the current signal size is greater than the signal size at the previous moment, the time point corresponding to the current signal is recorded as ST. When searching backward from ST on the curve SL for the first point that satisfies the current signal being greater than the signal at the next moment, the time point corresponding to the searched signal is recorded as ET. The above operation is repeated until all STs and ETs are marked. Starting from the first ST, each ST and the adjacent ET are respectively formed into a valid pulse group. The time interval between ST and ET of each valid pulse group is calculated in turn as the signal duration TIM of the group. With n as the serial number, TIM n Indicates the signal duration of the nth valid pulse group; Traverse all TIMs within the range of n. When TIM satisfies the formula Ln(TIM n )<ε, mark the valid pulse group corresponding to the current TIM as an abnormal pulse group, where ε is an arbitrary positive number, and traverse all abnormal pulse groups. If the formula Ln(TIM n )+ Ln(TIM n+1 )<ε or satisfy the formula Ln(TIM n )+ Ln(TIM n-1)<ε, the current abnormal pulse group is marked as a repeated pulse group, otherwise the current abnormal pulse group is marked as a lost pulse group; The statistical flow of people entering the store is corrected according to the abnormal pulse group. The specific method is: the flow of people entering the store plus the number of lost pulse groups minus the number of repeated pulse groups is used as the corrected flow of people entering the store.

[0035] Embodiment 2: Embodiment 2 replaces the method of calculating the mobile target's store-entry intention based on behavior information in Embodiment 1, specifically: In the video, the crowd density calculation algorithm is used to mark multiple groups of people to form a set GP, and j is used as the sequence number of the group. j Indicates the jth group of people. In the monitoring target set MOT, the curve ATY formed by the action trajectory of each moving target is obtained in turn. The curve ATY is a curve composed of the modulus of the time and the velocity vector. The velocity vector of the moving target is obtained as the target vector with a period of 2ms. At the same time, the velocity vectors of all people are obtained as the influence vector to form an influence set. Furthermore, the MCNN crowd density calculation algorithm is used to mark multiple crowds.

[0036] Traverse all velocity vectors on the curve ATY. If the velocity vector modulus at the current moment is smaller than the velocity vector modulus at the previous moment and also smaller than the velocity vector modulus at the next moment, or if the velocity vector modulus at the current moment is larger than the velocity vector modulus at the previous moment and also larger than the velocity vector modulus at the next moment, then mark the velocity vector at this moment as an affected vector. In the impact set, subtract all affected vectors from the impact vector in turn to find the magnitude of the vector modulus. Record the crowd corresponding to the impact vector with the largest modulus as the main impact group. Calculate the distance between the geometric center of the main impact group and the current moving target as the reference moment at the current time point. Calculate the distance between the geometric center of all crowds and the moving target and the magnitude of the reference moment. If it is larger than the reference moment, then record the current crowd as a weak impact group, otherwise, record it as a strong impact group. Traverse all the marked crowds in turn. If it is a strong influence group, subtract the velocity vector of the moving target at the current moment from the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. If it is a weak influence group, add the target vector of the moving target at the current moment to the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. The ratio of rat_h to the distance between the moving target and the store (the numerator of the calculated ratio is rat_h, and the denominator of the calculated ratio is the distance between the moving target and the store) is used as the local store-entering intention of the crowd for the moving target at the current moment. According to the above method, the local store-entering intention of the crowd for the moving target at all moments is calculated, and the sum of all local store-entering intentions is calculated as the store-entering intention of the current moving target.

[0037] In addition, the present invention also provides an embodiment of an intelligent store management system based on big data, such as Figure 2 Shown is a structural diagram of a smart store management system based on big data of the present invention. The smart store management system based on big data of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the smart store management system based on big data are implemented.

[0038] The system comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to run in the following units of the system: A real-time video recording unit is used to record surveillance videos outside the store through cameras outside the store; A behavior information acquisition unit, used to acquire behavior information of a moving target in the captured surveillance video outside the store; A fuzzy coefficient calculation unit, used for calculating the fuzzy coefficient of the moving target according to the acquired behavior information; A fuzzy information marking unit, used for marking the fuzzy information in the behavior information according to the calculated fuzzy coefficient; A store entry intention calculation unit, used to calculate the store entry intention of the mobile target according to the fuzzy information of the mobile target; The store flow counting unit is used to correct the current store flow according to the store entry intention of the mobile target.

[0039] The intelligent store management system based on big data can be run on computing devices such as desktop computers, notebooks, PDAs and cloud servers. The intelligent store management system based on big data can run on systems including, but not limited to, processors and memories. Those skilled in the art can understand that the example is only an example of an intelligent store management system based on big data, and does not constitute a limitation on an intelligent store management system based on big data. It can include more or fewer components than the example, or combine certain components, or different components. For example, the intelligent store management system based on big data can also include input and output devices, network access devices, buses, etc.

[0040] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the intelligent store management system based on big data, and uses various interfaces and lines to connect the various parts of the intelligent store management system based on big data.

[0041] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the intelligent store management system based on big data by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0042] Although the description of the present invention has been quite detailed and has been described in particular with respect to several described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, so as to effectively cover the intended scope of the present invention. In addition, the present invention is described above with the embodiments foreseeable by the inventors, and its purpose is to provide a useful description, and those non-substantial changes to the present invention that are not currently foreseen may still represent equivalent changes of the present invention.

Claims

1. A smart store management method based on big data, characterized in that: The method comprises the following steps: S100, obtaining behavior information of a moving target in a surveillance video outside a store; S200, calculating the fuzzy coefficient of the moving target based on the behavior information; S300, marking fuzzy information in the behavior information according to the fuzzy coefficient; S400, calculating the store entry intention of the mobile target according to the fuzzy information; S500: Correct the count of the flow of people entering the store based on the store entry intention.

2. The smart store management method based on big data according to claim 1 is characterized in that: In S100, the method for obtaining the behavior information of the mobile target in the surveillance video outside the store is: a plurality of mobile targets are marked in the video captured by the high-definition camera installed outside the store to form a monitoring target set MOT, and i is used as the sequence number of the mobile target, mot i is the i-th moving target; The video processing software Tracker is used to extract the speed and trajectory of the moving target, and the speed of all moving targets is formed into a speed set SPD, spd i is the speed of the ith moving target, calculate the distance between the moving target and the geometric center of the area occupied by the store, and the distances between all moving targets and the geometric center of the area occupied by the store form a distance set DST, dst i is the distance between the i-th moving target and the geometric center of the area occupied by the store. The speed set SPD and the distance set DST together constitute the behavior information of the moving target.

3. The smart store management method based on big data according to claim 2 is characterized in that: In S200, the method for calculating the fuzzy coefficient of the moving target based on the behavior information is: In the speed set, the speed of all moving targets is decomposed. The specific method is: The direction from the target to the geometric center of the area where the store is located is the horizontal direction, and the direction perpendicular to the geometric center of the area where the store is located is the vertical direction. The projection of the speed in the horizontal direction is spd_xi, and the projection of the speed in the vertical direction is spd_yi; Set the projection coefficient to m, and set the initial value of m to 1, calculate the difference between spd_xi and spd_yi, and modify the projection coefficient m according to the difference result; The video is divided into multiple video images according to the preset time t, and the distance between the moving target and the geometric center of the area occupied by the store is obtained at the same time. The distance of all moving targets and the corresponding acquisition time are plotted to generate a distance change curve XL, where the distance change curve XL is a curve composed of time and distance. The point on the XL curve is called the distance point dip, and k is the serial number of the acquisition time. dip k For T k The distance between the first target obtained at the moment and the geometric center of the area occupied by the store, where T k Indicates the kth acquisition time; Set the distance coefficient to x, and set the initial value of x to 1, calculate the sum of the derivatives at all distance points on the curve XL, and modify the distance coefficient x according to the calculation result; The speed of the moving target is obtained in the same way. The speed of all moving targets is plotted against the acquisition time to generate a speed change curve VL. The point on the curve VL is called a speed point rat. Let k be the serial number of the acquisition time, rat k For T k The speed of the moving target obtained at all times; Set the speed coefficient to v, and set the initial value of v to 1, calculate the sum of the derivatives at all speed points on the curve VL, and modify the speed coefficient v according to the calculation result; The fuzzy coefficient of the moving target is calculated based on the projection coefficient m, distance coefficient x and speed coefficient v.

4. The method for intelligent store management based on big data according to claim 3 is characterized in that: In S300, the method of marking fuzzy information in behavior information according to fuzzy coefficient is: all moving targets with fuzzy coefficient equal to zero are recorded as relevant targets, all video images containing relevant targets constitute a behavior set, and the behavior information of the relevant targets is recorded as fuzzy information.

5. The method for intelligent store management based on big data according to claim 4 is characterized in that: In S400, the method for calculating the mobile target's intention to enter the store based on the fuzzy information is: record the speed rat and distance dip corresponding to the relevant target collected on the curve VL and the curve XL at the latest moment, and calculate the current relevant target's intention to enter the store based on the speed rat and the distance dip.

6. The method for intelligent store management based on big data according to claim 5, characterized in that: The method of calculating the mobile target's intention to enter the store based on the behavior information is replaced by: using the crowd density calculation algorithm in the video to mark multiple groups of people to form a set GP, let j be the sequence number of the group, gp j Indicates the jth group of people. In the monitoring target set MOT, the curve ATY formed by the action trajectory of each moving target is obtained in turn, where the curve ATY is a curve composed of the modulus of time and velocity vector. The velocity vector of the moving target is obtained as the target vector according to the preset period t, and the velocity vectors of all people are obtained as the influence vector to form an influence set; Traverse all velocity vectors on the curve ATY. If the velocity vector modulus at the current moment is smaller than the velocity vector modulus at the previous moment and also smaller than the velocity vector modulus at the next moment, or if the velocity vector modulus at the current moment is larger than the velocity vector modulus at the previous moment and also larger than the velocity vector modulus at the next moment, then mark the velocity vector at this moment as an affected vector. In the impact set, subtract all affected vectors from the impact vector in turn to find the magnitude of the vector modulus. Record the crowd corresponding to the impact vector with the largest modulus as the main impact group. Calculate the distance between the geometric center of the main impact group and the current moving target as the reference moment at the current time point. Calculate the distance between the geometric center of all crowds and the moving target and the magnitude of the reference moment. If it is larger than the reference moment, then record the current crowd as a weak impact group, otherwise, record it as a strong impact group. Traverse all the marked crowds in turn. If it is a strong influence group, subtract the velocity vector of the moving target at the current moment from the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. If it is a weak influence group, add the target vector of the moving target at the current moment to the influence vector corresponding to the crowd, and then record the projection of the calculation result in the vertical direction as rat_h. The ratio of rat_h to the distance between the moving target and the geometric center of the area occupied by the store is taken as the local store-entering intention of the crowd for the moving target at the current moment. According to the above method, the local store-entering intention of the crowd for the moving target at all moments is calculated, and the sum of all local store-entering intentions is calculated as the store-entering intention of the current moving target.

7. The method for intelligent store management based on big data according to claim 6, characterized in that: In S500, the method for correcting the counting of the flow of people entering the store based on the intention of entering the store is: The pulse signal generated by the sensor is obtained at the time interval T within the preset period, and the generation time of the effective pulse is recorded as TIM_1, and the corresponding effective pulse disappearance time is recorded as TIM_2. A variable M is set, and the initial value of M is set to 0. Whenever a complete set of TIM_1 and TIM_2 is detected, or whenever a complete set of pulse signals is detected, the variable M is increased by 1. After the preset period ends, the value of the variable M is accessed as the store traffic during this period; The acquired pulse signal is plotted as a square wave signal SL, where the square wave signal SL is a curve composed of time and pulse signal. The size of the pulse signal is compared on the curve SL starting from the second acquisition time. If the current signal size is greater than the signal size at the previous moment, the time point corresponding to the current signal is recorded as ST. On the curve SL, search backward from ST for the first point that satisfies the current signal being greater than the signal at the next moment, and the time point corresponding to the searched signal is recorded as ET. Repeat the operation until all STs and ETs are marked. Starting from the first ST, each ST and the adjacent ET are respectively formed into a valid pulse group. The time interval between ST and ET of each valid pulse group is calculated in turn as the signal duration TIM of the group. With n as the serial number, TIM n Indicates the signal duration of the nth valid pulse group; After traversing all TIMs, mark the lost pulse groups and repeated pulse groups in all valid pulse groups, and then add the number of lost pulse groups to the store flow and subtract the number of repeated pulse groups to get the corrected store flow.

8. An intelligent store management system based on big data, characterized in that: The smart store management system based on big data includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the smart store management method based on big data described in any one of claims 1 to 7 are implemented.

Images