Deep neural network-based crowd concentration real-time evaluation method

Abstract

The invention discloses a crowd concentration degree real-time evaluation method based on a deep neural network. The crowd concentration degree real-time evaluation method comprises the following steps: acquiring a video containing crowds, and performing face detection Fi and storage; calling an algorithm GF for each Fi to obtain an attention signal Li (t); counting the standard deviation sigma i of the Li (t) to obtain the concentration degree sigma mean of one group; therefore, objective evaluation of attention levels of multiple persons is realized by adopting one camera.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning, and specifically is a real-time evaluation method for crowd concentration based on deep neural network. Background Art

[0002] Attention is a human psychological activity that is directed and focused on a certain object; attention is the ability to direct and focus psychological activities on a certain object.

[0003] Methods for attention detection include questionnaires, EEG-based methods, and virtual reality-based methods. These methods are either not quantitative, or have high or complex hardware requirements. There is no non-invasive, objective, and real-time inspection method that is not affected by head posture.

[0004] Currently, in some large communication venues, such as classrooms or conferences, the presenter's teaching level affects the quality of the entire report, and its evaluation is often subjective. Summary of the invention

[0005] The purpose of the present invention is to provide an objective detection method, which uses a camera to achieve objective evaluation of the attention level of multiple people, and provides an accurate evaluation of the lecturer's teaching level through analysis of the attention level of the crowd.

[0006] To achieve the above object, the present invention provides a method for real-time evaluation of crowd concentration based on a deep neural network, which is characterized by comprising the following steps:

[0007] Step 1: A video collection containing a group of people and face detection F i and store;

[0008] Step 2: For each F i Call algorithm G_F to get the attention signal L i (t);

[0009] Step 3: Count L i The standard deviation of (t) i , we get the concentration level of a group σ mean .

[0010] The method for real-time evaluation of crowd concentration based on deep neural network according to claim 1 is characterized in that in step 1, a video collection including group video, face detection F i And store, specifically adopt the following steps:

[0011] Step 1: Detect each person’s face F from an input video containing a group of people I(t) i ;

[0012] Step 2: i Stored as a list {F1,F2,F3,…}.

[0013] According to the method for real-time evaluation of crowd concentration based on deep neural network in claim 1, it is characterized in that in step 2, for each F i Call algorithm G_F to get the attention signal L i (t), the specific formula is as follows:

[0014] L i (t) = G_F(F i )

[0015] Among them, G_F is an attention detection method based on deep learning.

[0016] According to the method for real-time evaluation of crowd concentration based on deep neural network in claim 1, it is characterized in that in step 3, L i The standard deviation of (t) i , we get the concentration level of a group σ mean , specifically the following steps:

[0017] Step 1: Count L every 30 seconds i The standard deviation of (t) i ;

[0018] Step 2: Use σ i to indicate each person’s attention level;

[0019] Step 3: Calculate the multi-person σ in the group i The mean value σ mean ;

[0020] Step 4: Use σ mean To measure the focus of a group. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the flow chart of this patent;

[0022] Figure 2 Figure 1 is a diagram of a device for attention detection;

[0023] Figure 3 It is a structural diagram of a deep neural network;

[0024] Figure 4 is the attention signal L(t);

[0025] Figure 5 For multi-person attention detection;

[0026] Figure 6 Pay attention to the signal for multiple people. DETAILED DESCRIPTION

[0027] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] A real-time evaluation method for crowd concentration based on deep neural network is implemented as follows: Figure 3 As shown, it includes a PC and a camera. The PC hardware environment is Intel(R) Core(TM) i9-9900K, NVIDIA GTX1080Ti, DDR416GB and 500GB hard disk; the software environment is Ubuntu16.04 operating system, Python3.7.4 and Pytorch1.4.0 deep learning framework. The frame rate of the camera is 20fps and the resolution is 1280*720.

[0029] The camera was placed at the upper edge of the computer monitor, and the participants looked at the screen in a natural sitting position, approximately 60 cm away from the screen.

[0030] A real-time evaluation method for crowd concentration based on deep neural network, the process is as follows Figure 1 As shown, the following steps are included:

[0031] Step 1: A video collection containing a group of people and face detection F i and store;

[0032] Step 2: For each F i Call algorithm G_F to get the attention signal L i (t);

[0033] Step 3: Count L i The standard deviation of (t) i , we get the concentration level of a group σ mean .

[0034] For step one, the following steps are specifically used:

[0035] (1) Detect each person’s face F from an input video containing a group of people I(t) i ,like Figure 5 As shown;

[0036] (2) These F i Stored as a list {F1,F2,F3,…}.

[0037] For step 2, the following formula is used:

[0038] L i (t) = G_F(F i )

[0039] Among them, G_F is as follows Figure 2 The DenseNet deep neural network shown. The attention signal L obtained for a single person i (t) Figure 3 As shown, the attention signal L obtained for multiple people i (t) Figure 6 shown.

[0040] For step three, use the following steps:

[0041] (1) Count L every 30 seconds i The standard deviation of (t) i ;

[0042] (2) Using σ i To indicate each person's attention level

[0043] (3) Calculate σ for multiple people in a group i The mean value σ mean ;

[0044] (4) Using σ mean To measure the focus of a group.

Claims

1. A real-time evaluation method for crowd concentration based on deep neural network, characterized in that: The following steps are involved: Step 1: Collect a video containing a crowd and detect faces i and store; Step 2: For each F i Call algorithm G_F to get the attention signal L i (t); Step 3: Count L i The standard deviation of (t) i , we get the concentration level of a group σ mean .

2. According to the method of claim 1, the method is characterized in that: In step 1, a video containing a crowd is collected and face detection F i And store, specifically adopt the following steps: Step 1: Detect each person’s face F from an input video I(t) containing a crowd. i ; Step 2: i Stored as a list {F1,F2,F3,…}.

3. According to the method of claim 1, the method is characterized in that: In step 2, for each F i Call algorithm G_F to get the attention signal L i (t), the specific formula is as follows: L i (t)=G_F(F i ) Among them, G_F is an attention detection method based on deep learning.

4. According to claim 1, a method for real-time evaluation of crowd concentration based on deep neural network is characterized in that: Step 3: Count L i The standard deviation of (t) i , we get the concentration level of a group σ mean , specifically the following steps: Step 1: Count L every 30 seconds i The standard deviation of (t) i ; Step 2: Use σ i to indicate each person’s attention level; Step 3: Calculate the multi-person σ in the group i The mean value σ mean ; Step 4: Use σ mean To measure the focus of a group.

Images