Artificial intelligence-based method for elderly people to pick up meals in the cafeteria

Abstract

The present invention provides a method for elderly people to pick up meals in a cafeteria based on artificial intelligence, including the following steps: S1. Capture individual facial information and individual changes through a sensing device; S2. Perform liveness detection on individual change information to determine whether the liveness is RGB; S3. Establish positioning information and send the information to the application layer; S4. Perform face recognition on facial information and output face matching information; S5. The face matching information is transmitted to the platform layer, and the platform layer provides corresponding meals to the person picking up meals according to the face matching information, and establishes a big data model according to the dining data of the elderly in the community. The present invention can accurately calculate the dining data of the elderly through a face recognition algorithm, including data such as information of the elderly dining, meal structure, dining time, dining age, etc., can improve recognition efficiency, and quickly authenticate the information of the elderly dining. On the one hand, it solves the problem of long meal collection time for the elderly, and on the other hand, it improves the operational capacity of the cafeteria.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a method for elderly people to pick up meals in a cafeteria based on artificial intelligence. Background Art

[0002] With the continuous development of Internet technology, users can use mobile phones and other user terminals to order meals online and pick up meals by themselves. After completing the order, the user will obtain meal pickup vouchers such as meal pickup QR code and meal pickup number, and pick up the meal by themselves with the meal pickup voucher. Taking the meal pickup voucher as a meal pickup QR code as an example, when picking up the meal, the user needs to show the meal pickup QR code, and the QR code scanning device scans and verifies the meal pickup QR code. After successful verification, the user can pick up the meal.

[0003] However, the above dining method is not friendly enough for the elderly who do not know how to use mobile payment and have not been exposed to artificial intelligence applications. Special personnel are needed to guide the elderly in operation. For special canteens for the elderly, the time to pick up meals is too long and the efficiency is slow. Summary of the invention

[0004] In order to make up for the deficiencies in the prior art, the present invention provides a method for elderly people to pick up meals in a cafeteria based on artificial intelligence, and the technical solution adopted is as follows:

[0005] The method for taking meals in the cafeteria for the elderly based on artificial intelligence includes the following steps:

[0006] S1. Capture individual facial information and individual changes through sensing devices;

[0007] S2. Perform liveness detection on individual change information to determine whether the liveness is RGB;

[0008] S3. Establish positioning information and send information to the application layer;

[0009] S4. Perform face recognition on facial information and output face matching information;

[0010] S5. The facial matching information is transmitted to the platform layer. The platform layer provides the corresponding meal to the person picking up the meal based on the facial matching information, and establishes a big data model based on the dining data of the elderly in the community.

[0011] Furthermore, in step S2, the steps of liveness detection are as follows:

[0012] S21. Image acquisition, by simulating the positional relationship of human eyes, obtaining two images with different perspectives;

[0013] S22. Feature point matching, finding corresponding feature points in the two images;

[0014] S23. Calculate the disparity, and calculate the disparity of each point according to the matched feature points;

[0015] S24. Performing three-dimensional reconstruction of the target object;

[0016] S25. Analyze the three-dimensional reconstruction results to determine whether the target object is alive.

[0017] Furthermore, in step S22, feature point matching is performed based on SIFT and SURF algorithms.

[0018] Further, in step S24, the three-dimensional model is established by assisting with the phase difference method, including the following steps:

[0019] S241. Modulate the laser signal. The sine wave modulates the laser signal at the transmitting end. Assuming the modulation frequency is f, the modulated signal can be expressed as Asin(2πft), where A is the amplitude and t is the time;

[0020] S242. Signal propagation, the laser signal is sent from the transmitter, reflected by the target object and returned to the receiver; the round-trip time is t, then the total propagation time is 2d / c, where d is the distance of the target object and c is the speed of light;

[0021] S243. Phase difference measurement, calculate the phase difference of the signal propagation received by the receiving end

[0022] Furthermore, in step S4, the steps of face recognition are as follows:

[0023] S41. Collect the facial contour information of the target object and establish a training data set X = {X1, X2, X3, X4...Xn};

[0024] S42. The collected data set is preliminarily processed and the overall scaling is fixed to a size of 416*416 pixels;

[0025] S43. For the collected data that is fuzzy, poorly lit, or has unclear features, establish an image preprocessing mechanism;

[0026] S44. Extract image features in the convolution layer, add a normalization layer and an activation function;

[0027] S45. Establish a loss function of YOLO3, and establish a function for the loss of coordinate error, confidence error and category probability error of the recognition bounding box;

[0028] S46. Improve model generalization ability;

[0029] S47. Use the stochastic gradient descent algorithm to minimize the loss function;

[0030] S48. Train the model, perform non-maximum suppression and threshold filtering, and obtain the final detection result.

[0031] Furthermore, in step S43, the image preprocessing mechanism includes the following steps:

[0032] S431. Establish a correction set x = {x1, x2, x3, x4...xn};

[0033] S432. Perform image denoising on the data of the correction set, and establish a non-local mean filter. Where w((x, y), (i, j)) is the weight function and C(x, y) is the normalization constant;

[0034] S433. Increase contrast by using adaptive histogram equalization to perform histogram equalization in a local area. enhanceed (x, y) = clip(cdf local (I(x,y)))*(L-1).

[0035] Furthermore, in step S45, the coordinate error calculation formula is:

[0036]

[0037] The confidence error calculation formula is:

[0038]

[0039] The formula for calculating the category probability error is:

[0040]

[0041] Compared with the prior art, the present invention has the following beneficial technical effects:

[0042] The present invention can accurately calculate the dining data of the elderly through the face recognition algorithm, including the information of the elderly dining, meal structure, dining time, dining age and other data, which can improve the recognition efficiency and quickly authenticate the information of the elderly dining. On the one hand, it solves the problem of long time for the elderly to pick up meals, and on the other hand, it improves the operating capacity of the canteen. DETAILED DESCRIPTION

[0043] Many specific details are described in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from the description, and those skilled in the art can make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0044] Example 1

[0045] The method for taking meals in the cafeteria for the elderly based on artificial intelligence includes the following steps:

[0046] S1. Capture individual facial information and individual changes through sensing devices; the sensing devices include photosensitive sensors, laser sensors and binocular lenses.

[0047] S2. Perform liveness detection on the individual change information to determine whether the liveness is RGB.

[0048] S3. Establish positioning information and send information to the application layer;

[0049] S4. Perform face recognition on facial information and output face matching information;

[0050] S5. The facial matching information is transmitted to the platform layer. The platform layer provides the corresponding meal to the person picking up the meal based on the facial matching information, and establishes a big data model based on the dining data of the elderly in the community.

[0051] In step S2, the steps of liveness detection are as follows:

[0052] S21. Image acquisition, simulating the positional relationship of human eyes through binocular lenses to obtain two images with different perspectives;

[0053] S22. Feature point matching: find the corresponding feature points in the two images; feature point matching is based on SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features) algorithms. The goal is to find the same feature points in the two images, and the relative positions of these points in space provide information about the shape of the target object.

[0054] Among them, the steps of SIFT (Scale Invariant Feature Transform) algorithm are as follows:

[0055] (1) Scale space construction

[0056] Construct a Gaussian pyramid and obtain images of different scales by performing Gaussian blur and smoothing on the original image multiple times;

[0057] The calculation formula is: L(x, y, σ) = G(x, y, σ) * I(x, y). Among them, G(x, y, σ) is a Gaussian kernel with a standard deviation of σ, I(x, y) is the original image, and * represents the convolution operation.

[0058] (2) Extreme point detection: Find extreme points in the scale space, that is, local maximum or minimum points at adjacent scales and spatial positions. By comparing a pixel with all 8 pixels in its neighborhood and the 9 pixels corresponding to the upper and lower layers, we can determine whether it is an extreme point.

[0059] (3) Key point positioning: Refine the detected extreme points with precise positions and scales, and remove edge responses and low-contrast points. Use second-order Taylor expansion for precise sub-pixel positioning.

[0060] (4) Direction assignment: assign one or more directions to each key point to enhance rotation invariance. By performing histogram statistics on the gradient directions of the area around the key point, the direction with the largest peak value is selected as the direction of the key point.

[0061] (5) Key point description: extract the feature vector of the area around the key point to form a descriptor. Usually a 16x16 window is used, divided into 4x4 blocks, and the gradient histogram in 8 directions is calculated for each block, eventually forming a 128-dimensional feature vector.

[0062] The SURF (Speeded Up Robust Features) algorithm is an improvement on SIFT, which aims to improve the speed and efficiency of feature detection. SURF simplifies the calculation by using the integral image and the approximation of the Hessian matrix while maintaining good feature detection performance. The algorithm steps are as follows:

[0063] (1) Integral image calculation

[0064] Pre-compute the integral image to quickly calculate the sum of pixels in any rectangular area.

[0065] The integral image II(x,y) is defined as the sum of all pixel values ​​from the top left corner of the image to the point (x,y).

[0066] (2) Approximation of the Hessian matrix

[0067] The second-order partial derivatives are approximated using a box filter and the Hessian matrix is ​​constructed.

[0068] The Hessian matrix is ​​defined as:

[0069] (3) Extreme point detection

[0070] Find the extreme points of the determinant of the Hessian matrix in scale space.

[0071] Calculate the determinant D = det(H) = L xx L yy -(L xy ) 2 , and find the local maximum point.

[0072] (4) Key point positioning

[0073] The detected extreme points are accurately located at the sub-pixel level, edge responses and low-contrast points are eliminated, and interpolation methods are used for refinement.

[0074] (5) Direction assignment

[0075] One or more directions are assigned to each key point to enhance rotation invariance. The direction with the largest peak is selected as the direction of the key point by performing histogram statistics on the Haar wavelet response of the area around the key point.

[0076] (6) Key point description

[0077] Extract the feature vector of the area around the key point to form a descriptor. Usually a 25x25 window is used, divided into 4x4 blocks, and the Haar wavelet response in 25 directions is calculated for each block, eventually forming a 64-dimensional feature vector.

[0078] S23. Calculate the disparity, and calculate the disparity of each point based on the matched feature points; the disparity refers to the position difference of the same feature point in two images. The size of the disparity is inversely proportional to the distance from the point to the camera, that is, the closer the distance, the greater the disparity.

[0079] S24. Perform three-dimensional reconstruction of the target object; use the parallax information and camera parameters (such as focal length, baseline length, etc.) to calculate the spatial coordinates of each feature point through triangulation to achieve three-dimensional reconstruction of the target object. This process can generate a three-dimensional point cloud or mesh model that reflects the true shape of the target object.

[0080] The three-dimensional model is established by phase difference method, including the following steps:

[0081] S241. Modulate the laser signal. The sine wave modulates the laser signal at the transmitting end. Assuming the modulation frequency is f, the modulated signal can be expressed as Asin(2πft), where A is the amplitude and t is the time;

[0082] S242. Signal propagation, the laser signal is sent from the transmitter, reflected by the target object and returned to the receiver; the round-trip time is t, then the total propagation time is 2d / c, where d is the distance of the target object and c is the speed of light;

[0083] S243. Phase difference measurement, calculate the phase difference of the signal propagation received by the receiving end

[0084] S25. Liveness detection, analyze the 3D reconstruction results, determine whether the target object is alive, and output the detection results to the application system for subsequent identity verification and other operations. Check whether there is a reasonable depth change. Real faces have obvious depth information, while flat images do not. The texture features of the faces of the elderly are larger, such as the fineness of the skin and pores. The texture of real skin is complex, and printed photos or images displayed on the screen often lack these details. In addition, through the reaction under different lighting conditions, the real face will show natural shadows and highlights according to the changes in lighting.

[0085] In step S4, the steps of face recognition are as follows:

[0086] S41. Collect the facial contour information of the target object and establish a training data set X = {X1, X2, X3, X4...Xn};

[0087] S42. The collected data set is preliminarily processed and the overall scaling is fixed to a size of 416*416 pixels;

[0088] S43. For the collected data that is fuzzy, poorly lit, or has unclear features, establish an image preprocessing mechanism; the image preprocessing mechanism includes the following steps:

[0089] S431. Establish a correction set x = {x1, x2, x3, x4...xn};

[0090] S432. Perform image denoising on the data of the correction set, and establish a non-local mean filter. Where w((x, y), (i, j)) is the weight function and C(x, y) is the normalization constant;

[0091] S433. Increase contrast by using adaptive histogram equalization to perform histogram equalization in a local area. enhanceed (x, y) = clip(cdf local (I(x,y)))*(L-1).

[0092] S44. Extract image features in the convolution layer, add a normalization layer and an activation function;

[0093] S45. Establish a loss function of YOLO3, and establish a function for the loss of coordinate error, confidence error and category probability error of the recognition bounding box;

[0094] Coordinate loss, using mean square error to calculate the error between the center coordinates of the convenience box and the center coordinates of the true bounding box:

[0095]

[0096] Confidence loss, using binary cross entropy loss to calculate the error between the predicted confidence and the true confidence:

[0097]

[0098] Category loss, using cross entropy loss to calculate the predicted category probability and the true category error:

[0099]

[0100] S46. Improve the generalization ability of the model, random cropping, flipping and color jittering, etc.;

[0101] S47. Use the stochastic gradient descent algorithm to minimize the loss function;

[0102] S48. Train the model, perform non-maximum suppression and threshold filtering, and obtain the final detection result.

[0103] The liveness detection of the present invention meets the actual needs. The liveness detection is tested by switching back and forth between photos and real people. The data shows that the liveness detection rate is 99.8% when the light is sufficient, and 95.3% in a dark environment. The algorithm model built from the image information captured by the liveness detection camera to the recognition of the face image meets the expected requirements. The recognition can be completed in 300-400ms, and the recognition accuracy is 99.7%.

[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to replace some of the technical features therein with equivalents. However, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for elderly people to pick up food in a cafeteria based on artificial intelligence, characterized in that: The following steps are involved: S1. Capture individual facial information and individual changes through sensing devices; S2. Perform liveness detection on individual change information to determine whether the liveness is RGB; S3. Establish positioning information and send information to the application layer; S4. Perform face recognition on facial information and output face matching information; S5. The facial matching information is transmitted to the platform layer. The platform layer provides the corresponding meal to the person picking up the meal based on the facial matching information, and establishes a big data model based on the dining data of the elderly in the community.

2. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to claim 1, characterized in that: In step S2, the steps of liveness detection are as follows: S21. Image acquisition, by simulating the positional relationship of human eyes, obtaining two images with different perspectives; S22. Feature point matching, finding corresponding feature points in the two images; S23. Calculate the disparity, and calculate the disparity of each point according to the matched feature points; S24. Performing three-dimensional reconstruction of the target object; S25. Analyze the three-dimensional reconstruction results to determine whether the target object is alive.

3. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to claim 2, characterized in that: In step S22, feature point matching is performed based on SIFT and SURF algorithms.

4. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to claim 3 is characterized in that: In step S24, a three-dimensional model is established by using a phase difference method, including the following steps: S241. Modulate the laser signal. The sine wave modulates the laser signal at the transmitting end. Assuming the modulation frequency is f, the modulated signal can be expressed as Asin(2πft), where A is the amplitude and t is the time; S242. Signal propagation, the laser signal is sent from the transmitter, reflected by the target object and returned to the receiver; the round-trip time is t, then the total propagation time is 2d / c, where d is the distance of the target object and c is the speed of light; S243. Phase difference measurement, calculate the phase difference of the signal propagation received by the receiving end 5. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to any one of claims 1 to 4, characterized in that: In step S4, the steps of face recognition are as follows: S41. Collect the facial contour information of the target object and establish a training data set X = {X1, X2, X3, X4...Xn}; S42. The collected data set is preliminarily processed and the overall scaling is fixed to a size of 416*416 pixels; S43. For the collected data that is fuzzy, poorly lit, or has unclear features, establish an image preprocessing mechanism; S44. Extract image features in the convolution layer, add a normalization layer and an activation function; S45. Establish a loss function of YOLP3, and establish a function for the loss of coordinate error, confidence error and category probability error of the recognition bounding box; S46. Improve model generalization ability; S47. Use the stochastic gradient descent algorithm to minimize the loss function; S48. Train the model, perform non-maximum suppression and threshold filtering, and obtain the final detection result.

6. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to claim 5, characterized in that: In step S43, the image preprocessing mechanism includes the following steps: S431. Establish a correction set x = {x1, x2, x3, x4...xn}; S432. Perform image denoising on the data of the correction set, and establish a non-local mean filter. Where w((x, y), (i, j)) is the weight function and C(x, y) is the normalization constant; S433. Increase contrast by using adaptive histogram equalization to perform histogram equalization in a local area. enhanceed (x, y) = clip(cdf local (I(x,y)))*(L-1).

7. The method for taking food from a cafeteria for the elderly based on artificial intelligence according to claim 6, characterized in that: In step S45, the coordinate error calculation formula is: The confidence error calculation formula is: The formula for calculating the category probability error is:

Images