A method and device for voice broadcasting of bus seats based on face age detection

Through the face age detection technology based on convolutional neural network, the elderly and young people are identified and reminded, the problem of occupancy of love seats in public transportation is solved, and the intelligent management of bus seats and the improvement of passenger experience is achieved.

CN114677724BActive Publication Date: 2024-11-08NANJING UNIV OF POSTS & TELECOMM
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210199578.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-02
Publication Date
2024-11-08
Estimated Expiration
2042-03-02

AI Technical Summary

Technical Problem

In public transportation, caring seats are often occupied by passengers at will, making it impossible for vulnerable groups who really need special seats to be used, and it is difficult for the existing technology to accurately identify and remind the elderly and young people.

Method used

Design a voice broadcast method for bus seats based on face age detection. Using convolutional neural network and facial recognition module, it collects passenger face videos in real time, recognizes and determines the elderly and young objects. If there are elderly and young people, voice prompts will be issued. , ensure the reasonable allocation of special seats.

Benefits of technology

It improves the efficiency and passenger experience of bus seats, realizes intelligent management of bus seats, ensures orderly rides for passengers and improves travel experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114677724B_ABST
    Figure CN114677724B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for voice broadcasting of special seats for public transportation based on face age detection, wherein the method is based on a facial recognition module, a face age detection module, and an old and young object detection module, takes a video of all facial parts of a target population collected in real time as input, takes the state of whether there are old and young objects in the target population as output, and constructs an old and young population detection model, the device comprises a first camera, a second camera, a first old and young population detection module, a second old and young population detection module, a voice broadcast module, and a main control module, and the present invention provides a reminder for giving up seats to the old and young age groups through intelligent management and control of special seats for public transportation, thereby preventing passengers from having disputes over giving up seats to a certain extent, facilitating passengers' transportation travel, and helping to create a harmonious riding environment and improve people's transportation travel experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of face detection and public transportation, and in particular to a method and device for voice broadcasting of special bus seats based on face age detection. Background Art

[0002] As the pace of life is getting faster and faster, traveling by public transportation has become an indispensable part of people's lives. Taking into account some special passengers, such as the elderly, children, and parents with children, public transportation such as buses and subways have begun to set up special seats for these vulnerable groups. However, due to the large passenger flow on buses, subways and other transportation vehicles and the maintenance of order depends solely on the self-consciousness of the passengers, the special seats in the car often cannot accurately serve the target population, and passengers often occupy the special seats at will, and when passengers who really need the special seats get on the bus, they cannot use them.

[0003] Face detection age technology is a machine-based deep learning technology based on artificial intelligence. It infers the approximate age or age range (age group) of the target object based on the facial image. Due to the influence of factors such as makeup, lighting, obstacles and facial expressions, it is difficult to accurately guess the exact age from a face image; therefore, the present invention treats it as a classification problem rather than a regression problem. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for voice broadcasting of special seats on buses based on face age detection, and to provide seat-giving reminders to people of different age groups based on the recognition of the age groups of target objects.

[0005] In order to achieve the above functions, the present invention designs a bus seat voice broadcast method based on face age detection, according to the following steps S1-S5, an old and young crowd detection model is obtained, and then the old and young crowd detection model is applied to complete the detection of old and young objects in the target crowd;

[0006] S1. Use a preset fixed image to obtain postures and collect videos including all facial parts of the target population in real time;

[0007] S2. Taking the video including all facial parts of the target population as input, constructing a facial recognition module based on image enhancement preprocessing and taking the local images of each facial part of a preset size as output;

[0008] S3. Taking the local images of each facial part as input, based on the convolutional neural network, according to the age group labels corresponding to each facial part according to the preset mapping rules, and taking the age group labels corresponding to all facial parts of the target population as output, a facial age detection module is constructed;

[0009] S4. Taking the age group labels corresponding to all facial parts of the target population as input, if there is an age group label higher than the preset upper limit or lower than the preset lower limit, it is determined that there are old and young objects in the target population, otherwise it is determined that there are no old and young objects in the target population, and the state of whether there are old and young objects in the target population is used as output to construct an old and young object detection module;

[0010] S5. Based on the facial recognition module, face age detection module, and old and young object detection module, a model for detecting old and young people is constructed by taking the real-time collected video of all facial parts of the target population as input and the status of whether there are old and young objects in the target population as output.

[0011] As a preferred technical solution of the present invention: the image enhancement preprocessing in step S2 includes random cropping and random rotation.

[0012] As a preferred technical solution of the present invention: it is characterized in that a padding layer is added to the convolutional neural network in step S3 to make up for the size difference of the video frames of each face part input to the face age detection module, so that the input dimension and output dimension of the convolutional layer of the convolutional neural network are consistent, specifically as follows:

[0013]

[0014]

[0015] In the formula, H out is the height of the output image, H in is the height of the input image, W out is the width of the output image, W in is the width of the input image, kernel size[m] is the size of the pooling kernel of each convolutional neural network, stride[m] is the length of the horizontal and vertical paths in the convolution process, the default value is 1, padding[m] indicates padding the edges of the input image, dilation[m] indicates a dilated convolution; m is 0 or 1, [0] is used to indicate the parameters on the input layer, and [1] is used to indicate the parameters on the output layer;

[0016] A local response normalization layer is added to the convolutional neural network to improve the generalization ability of the convolutional neural network, as shown in the following formula:

[0017]

[0018] In the formula, represents the value of the input local response normalization layer, is the value output after being processed by the local response normalization layer, a is the preset channel parameter, i is the channel position, and j is The maximum value in , x, y are the horizontal and vertical coordinates of the pixel to be updated, k is the preset bias, n is the channel number, N is the total number of channels, α, β are the preset hyperparameters.

[0019] As a preferred technical solution of the invention: the age group labels in step S3 are 8 in total, namely (0-2), (4-6), (8-12), (15-20), (25-32), (38-43), (48-53), and (60-100). As a preferred technical solution of the invention: it also includes a control panel, which is installed on the bicycle frame and is connected to the main control module and the power module respectively.

[0020] As a preferred technical solution of the invention: it is characterized in that in step S3, the convolutional neural network is made to converge based on the stochastic gradient descent method, which is specifically as follows:

[0021]

[0022] In the formula, J (i) (θ0, θ1) is the loss function, h θ (x i ) is the prediction function, θ is the derivative angle;

[0023] To J (i) (θ0, θ1) is derived:

[0024]

[0025] In the formula, θ j is the objective function;

[0026] For the objective function θ j Update the parameters:

[0027]

[0028] In the formula, θ′ j is the objective function after updating the parameters, α0 is the learning rate, and θ j Assign initial values, and then use J (i) (θ0, θ1) updates the objective function θ in the direction with the largest change j until the preset maximum number of iterations.

[0029] The present invention also provides a bus seat voice broadcast device based on face age detection, which is characterized by comprising a first camera, at least one second camera, a first old and young group detection module, at least one second old and young group detection module, a voice broadcast module, and a main control module, so that the bus seat voice broadcast device based on face age detection can realize the bus seat voice broadcast method based on face age detection;

[0030] Among them, the number of second cameras and second old and young group detection modules is the same. The first camera for real-time collection of videos of all facial parts of passengers getting on the bus is fixedly installed at the bus door. Each second camera for real-time collection of videos of facial parts of passengers using special bus seats corresponds to each special bus seat and is fixedly installed at each special bus seat. The first camera and the main control module are respectively connected to the first old and young group detection module, and each second camera is respectively connected to each second old and young group detection module in a one-to-one correspondence. Each second old and young group detection module is connected to the main control module, and the main control module is connected to the voice broadcast module. The first old and young group detection module and each second old and young group detection module respectively include a facial recognition module, a face age detection module, and an old and young object detection module. When the first old and young group detection module detects that there are old and young objects among the passengers getting on the bus, and each second old and young group detection module detects that there are no old and young objects among the passengers using special bus seats, the main control module drives the voice broadcast module to emit a preset voice prompt sound.

[0031] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0032] The present invention designs a method and device for voice broadcasting of special bus seats based on face age detection. The face detection age technology based on convolutional neural network further improves the utilization efficiency of special bus seats, realizes intelligent management of special bus seats, facilitates passengers to take the bus in a more orderly manner, and improves the passengers' bus travel experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart of a method for voice broadcasting of bus seats based on face age detection according to an embodiment of the present invention;

[0034] Figure 2 is a confusion matrix diagram of the accuracy of age group labels output by the face age detection module provided by an embodiment of the present invention;

[0035] Figure 3a to Figure 3c 1 is a loss curve of a convolutional neural network training model and a schematic diagram of the accuracy of a validation set according to an embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of a bus seat voice announcement device based on face age detection provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are not intended to limit the scope of protection of the present invention. Figure 1The method for voice broadcasting of bus seats based on face age detection provided by the embodiment of the present invention obtains an old and young crowd detection model according to the following steps S1 to S5, and then uses the old and young crowd detection model to complete the detection of old and young objects in the target crowd;

[0038] S1. Use a preset fixed image to obtain postures and collect videos including all facial parts of the target population in real time;

[0039] S2. Taking a video including all facial parts of the target population as input, based on image enhancement preprocessing, and taking local images of each facial part of a preset size as output, a facial recognition module is constructed; the image enhancement preprocessing includes random cropping and random rotation.

[0040] S3. Taking the local images of each facial part as input, based on the convolutional neural network, according to the age group labels corresponding to each facial part according to the preset mapping rules, and taking the age group labels corresponding to all facial parts of the target population as output, a facial age detection module is constructed;

[0041] The convolutional neural network contains 3 convolutional layers and 2 fully connected layers. Finally, it is divided into 8 categories by SVM classifier. The output layer type is softmax type. The padding layer is added to the convolutional neural network to make up for the size difference of the video frames of each face part input to the face age detection module, so that the input dimension and output dimension of the convolutional layer of the convolutional neural network are consistent. The specific formula is as follows:

[0042]

[0043]

[0044] In the formula, H out is the height of the output image, H in is the height of the input image, W out is the width of the output image, W in is the width of the input image, kernel size[m] is the size of the pooling kernel of each convolutional neural network, stride[m] is the length of the horizontal and vertical paths in the convolution process, the default value is 1, padding[m] indicates padding the edges of the input image, dilation[m] indicates a dilated convolution; m is 0 or 1, [0] is used to indicate the parameters on the input layer, and [1] is used to indicate the parameters on the output layer;

[0045] A local response normalization layer is added to the convolutional neural network to improve the generalization ability of the convolutional neural network, as shown in the following formula:

[0046]

[0047] In the formula, represents the value of the input local response normalization layer, is the value output after being processed by the local response normalization layer, a is the preset channel parameter, i is the channel position, and j is The maximum value in , x, y are the horizontal and vertical coordinates of the pixel to be updated, k is the preset bias, n is the channel number, N is the total number of channels, α, β are the preset hyperparameters.

[0048] In one embodiment, SVM classifier is used for classification, and there are 8 age group labels, namely (0-2), (4-6), (8-12), (15-20), (25-32), (38-43), (48-53), and (60-100); the confusion matrix diagram of the accuracy of the age group labels output by the face age detection module provided in the embodiment of the present invention is referenced Figure 2 The rows of the confusion matrix in the figure represent the number of age groups that each face actually belongs to, and the columns represent the number of faces classified as this age group by the convolutional neural network. The confusion matrix is ​​a standard format for expressing accuracy evaluation.

[0049] The convolutional neural network is converged based on the stochastic gradient descent method, as shown in the following formula:

[0050]

[0051] In the formula, J (i) (θ0, θ1) is the loss function, h θ (x i ) is the prediction function, θ is the derivative angle;

[0052] To J (i) (θ0, θ1) is derived:

[0053]

[0054] In the formula, θ j is the objective function;

[0055] For the objective function θ j Update the parameters:

[0056]

[0057] In the formula, θ′ j is the objective function after updating the parameters, α0 is the learning rate, and θ j Assign initial values, and then use J (i) (θ0, θ1) updates the objective function θ in the direction with the largest change j until the preset maximum number of iterations.

[0058] S4. Taking the age group labels corresponding to all facial parts of the target population as input, if there is an age group label higher than the preset upper limit or lower than the preset lower limit, it is determined that there are old and young objects in the target population, otherwise it is determined that there are no old and young objects in the target population, and the state of whether there are old and young objects in the target population is used as output to construct an old and young object detection module;

[0059] S5. Based on the facial recognition module, face age detection module, and old and young object detection module, a model for detecting old and young people is constructed by taking the real-time collected video of all facial parts of the target population as input and the status of whether there are old and young objects in the target population as output.

[0060] In one embodiment, the face age detection module is based on the tensorflow model, which has fast running speed and good real-time performance. The transplantation platform uses the Raspberry Pi 4b; the convolutional neural network is based on the Alexnet model, and the drop out value used to limit the overfitting of the convolutional neural network is 0.5. The number of samples selected before the convolutional neural network adjusts the parameters each time is 50, and the number of parameters passed to the program for training at a single time is 256; Figure 3a The training data set loss curve of the convolutional neural network training model provided in the embodiment of the present invention is Figure 3b This is the validation set loss curve for the convolutional neural network training model. Figure 3c is the validation set accuracy of the convolutional neural network training model, Figure 3c It shows that after more than 5 iterations, the accuracy of the validation set reaches more than 90%, which can achieve accurate recognition of face age.

[0061] The bus seat voice announcement device based on face age detection provided by the embodiment of the present invention comprises a first camera, at least one second camera, a first old and young group detection module, at least one second old and young group detection module, a voice announcement module, and a main control module, so that the bus seat voice announcement device based on face age detection can realize the bus seat voice announcement method based on face age detection;

[0062] refer to Figure 4, wherein the number of second cameras and second old and young group detection modules is the same, the first camera for real-time collection of all facial videos of boarding passengers is fixedly installed at the bus door, each second camera for real-time collection of facial videos of passengers using special bus seats corresponds to each special bus seat, and is fixedly installed at each special bus seat, the first camera and the main control module are respectively connected to the first old and young group detection module, each second camera is respectively connected to each second old and young group detection module in a one-to-one correspondence, each second old and young group detection module is connected to the main control module, the main control module is connected to the voice broadcast module, the first old and young group detection module and each second old and young group detection module respectively include a facial recognition module, a facial age detection module, and an old and young object detection module, when the first old and young group detection module detects that there are old and young objects among the boarding passengers, and each second old and young group detection module detects that there are no old and young objects among the passengers using special bus seats, the main control module drives the voice broadcast module to emit a preset voice prompt sound.

[0063] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A bus seat voice announcement method based on face age detection, It is characterized in that According to the following steps S1 to S5, an old and young population detection model is obtained, and then the old and young population detection model is applied to complete the detection of old and young objects in the target population; S1. Use a preset fixed image to obtain postures and collect videos including all facial parts of the target population in real time; S2. Taking the video including all facial parts of the target population as input, constructing a facial recognition module based on image enhancement preprocessing and taking the local images of each facial part of a preset size as output; S3. Taking the local images of each facial part as input, based on the convolutional neural network, according to the age group labels corresponding to each facial part according to the preset mapping rules, and taking the age group labels corresponding to all facial parts of the target population as output, a facial age detection module is constructed; A padding layer is added to the convolutional neural network in step S3 to make up for the size difference of the video frames of each face part input to the face age detection module, so that the input dimension and output dimension of the convolutional layer of the convolutional neural network are consistent, as shown in the following formula: In the formula, H out is the height of the output image, H in is the height of the input image, W out is the width of the output image, W in is the width of the input image, kernel size[m] is the size of the pooling kernel of each convolutional neural network, stride[m] is the length of the horizontal and vertical paths during the convolution process, the default value is 1, padding[m] indicates padding the edges of the input image, and dilation[m] indicates a dilated convolution; m is 0 or 1, [0] is used to indicate the parameters on the input layer, and [1] is used to indicate the parameters on the output layer; A local response normalization layer is added to the convolutional neural network to improve the generalization ability of the convolutional neural network, as shown in the following formula: In the formula, represents the value of the input local response normalization layer, is the value output after the local response normalization layer is processed, α is the preset channel parameter, i is the channel position, and j is The maximum value in , x, y are the horizontal and vertical coordinates of the pixel to be updated, k is the preset bias, n is the channel number, N is the total number of channels, α, β are the preset hyperparameters; S4. Taking the age group labels corresponding to all facial parts of the target population as input, if there is an age group label higher than the preset upper limit or lower than the preset lower limit, it is determined that there are old and young objects in the target population, otherwise it is determined that there are no old and young objects in the target population, and the state of whether there are old and young objects in the target population is used as output to construct an old and young object detection module; S5. Based on the facial recognition module, face age detection module, and old and young object detection module, a model for detecting old and young people is constructed by taking the real-time collected video of all facial parts of the target population as input and the status of whether there are old and young objects in the target population as output.

2. A method for voice broadcasting bus seats based on face age detection as claimed in claim 1, It is characterized in that The image enhancement preprocessing in step S2 includes random cropping and random rotation.

3. A bus seat voice announcement method based on face age detection as claimed in claim 1, It is characterized in that There are 8 age group labels in step S3, namely (0-2), (4-6), (8-12), (15-20), (25-32), (38-43), (48-53), and (60-100).

4. A method for voice broadcasting bus seats based on face age detection as claimed in claim 3, It is characterized in that In step S3, the convolutional neural network is converged based on the stochastic gradient descent method, which is specifically as follows: In the formula, J (i) (θ 0 ,θ 1 ) is the loss function, h θ (x i ) is the prediction function, θ is the derivative angle; To J (i) (θ 0 ,θ 1 ) to find the derivative: In the formula, θ j is the objective function; For the objective function θ j Update the parameters: In the formula, θ′ j is the objective function after updating the parameters, α 0 is the learning rate, and θ j Assign initial values, and then use J (i) (θ 0 ,θ 1 ) Update the objective function θ in the direction with the largest change j until the preset maximum number of iterations.

5. A bus seat voice announcement device based on face age detection, It is characterized in that The device comprises a first camera, at least one second camera, a first old-child group detection module, at least one second old-child group detection module, a voice broadcast module, and a main control module, so that the voice broadcast device for special bus seats based on face age detection can realize the voice broadcast method for special bus seats based on face age detection as claimed in any one of claims 1 to 4; Among them, the number of second cameras and second old and young group detection modules is the same. The first camera for real-time collection of videos of all facial parts of passengers getting on the bus is fixedly installed at the bus door. Each second camera for real-time collection of videos of facial parts of passengers using special bus seats corresponds to each special bus seat and is fixedly installed at each special bus seat. The first camera and the main control module are respectively connected to the first old and young group detection module, and each second camera is respectively connected to each second old and young group detection module in a one-to-one correspondence. Each second old and young group detection module is connected to the main control module, and the main control module is connected to the voice broadcast module. The first old and young group detection module and each second old and young group detection module respectively include a facial recognition module, a face age detection module, and an old and young object detection module. When the first old and young group detection module detects that there are old and young objects among the passengers getting on the bus, and each second old and young group detection module detects that there are no old and young objects among the passengers using special bus seats, the main control module drives the voice broadcast module to emit a preset voice prompt sound.

Citation Information

Patent Citations

  • Face image age estimation method, device and terminal equipment

    CN107169454A

  • Method and device for identifying ages of population

    CN107545249A

  • Gender and age recognition method and device and storage medium

    CN112257503A

  • Training method of age prediction model, age prediction method and related device

    CN113076823A