A human body falling recognition algorithm based on 3D vision
Through the human body dump recognition algorithm based on 3D vision, light image sensors and infrared sensors are combined with Sobel operator to detect human body feature points, identifying the state of being about to fall, solving the problem of difficulty in time for middle-aged and elderly people to treat when falling, and improving safety sex.
Patent Information
- Application Number
- CN202210188837.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-28
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-02-28
AI Technical Summary
It is difficult for middle-aged and elderly people to receive timely treatment when falling, resulting in potential dangers and injuries. It is difficult for the existing technology to efficiently identify and prevent human penetration behavior.
Using a human body dump recognition algorithm based on 3D vision, the human body moving images are collected through the light image sensor, infrared transmitter, infrared receiver and camera sensor module, detect feature points and calculate the angle between the center of gravity and the ground, and determine whether it is about to be dumped. Combined with the Sobel operator and distance detection improve recognition efficiency.
It realizes convenient identification of characters about to fall, improves the efficiency of identifying behaviors, avoids the occurrence of fall, and ensures the safety of the elderly.
Smart Images

Figure CN114565974B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of 3D vision, and in particular to a human body falling recognition algorithm based on 3D vision. Background Art
[0002] As China's aging trend intensifies, the number of middle-aged and elderly people is increasing. At the same time, as people's safety awareness increases, how to ensure personal safety and how to promptly detect people in dangerous conditions are receiving more and more attention.
[0003] The middle-aged and elderly are the most vulnerable group to falls, especially those who have suffered a stroke. Before a stroke occurs, they may experience symptoms such as dizziness, headache, and blackouts, which can easily lead to a sudden fall. The consequences of a sudden fall caused by a stroke without quick treatment are unimaginable. In addition, some people with limited mobility may suffer from incurable injuries if they cannot receive quick and timely treatment after an accidental fall. How to detect falls in the elderly in a timely manner is receiving more and more attention. Summary of the invention
[0004] In view of the above technical problems, the present invention provides a human body falling recognition algorithm based on 3D vision, which can conveniently identify that a human target is about to fall, improve the efficiency of identifying falling behavior, and avoid the occurrence of falling.
[0005] To achieve the above object, the present invention provides the following technical solution: a 3D vision-based human body tilting recognition algorithm, comprising the following steps:
[0006] Step 1: collecting human body moving images, wherein an optical image sensor, an infrared transmitter, an infrared receiver, and a camera sensor module are used to collect human body moving images, and human face recognition is performed to obtain the human body standing face orientation, wherein the camera sensor module includes an RGB image information component and a depth information component;
[0007] Step 2: Obtain the image information collected in step 1, and detect and separate the image objects in the image information;
[0008] Step 3: Tracking the feature points on the human target model separated in step 2, wherein the collection positions of the feature points on the human target model include the head, neck, chest and legs;
[0009] Step 4: Identify the tendency of the target through the acquired feature points;
[0010] In step 4, the process of identifying the tendency of the target using the acquired feature points is as follows:
[0011] Step 1: Obtain the position of feature points in space at a certain moment through the two-dimensional information and depth information of the human body:
[0012] Step 4: According to the first to third steps, the vector between the legs and the chest can be obtained to determine whether the target is in a moving state, and then the angle between the straight line where the head and leg feature points are located and the vertical direction is calculated;
[0013] Step 5: Obtain the target inclination direction through the target inclination recognition information, and determine the state of the target, where the human body movement state includes normal upright movement state, about to fall state, and falling state. The about to fall state is determined by the angle between the target center of gravity and the ground.
[0014] Preferably, in step 2, the moving target is detected by detecting the changed area, and the first-order image gradient information is obtained by combining the Sobel operator to obtain target segmentation, and the information after the target human body detected in the image is separated from the background is obtained.
[0015] Preferably, the step 2 also includes distance detection, and the distance detection obtains an RGB image through the camera sensor module and inputs it into a SoC sensor processor to obtain the distance between the image target and the camera.
[0016] Preferably, the state of about to fall described in step five includes a state of the target person leaning forward and backward and a state of the target person leaning left and right, wherein the state of the target person leaning forward and backward is when the angle between the target's forward and backward inclination direction and the center of gravity direction is greater than 12.5°; the state of the target person leaning left and right is when the angle between the target's left and right inclination direction and the center of gravity direction is greater than 16°.
[0017] The beneficial effects of the present invention are as follows: by collecting the moving image of the target person, it is convenient to determine whether the target person is in motion, so as to further calculate whether the target person is about to fall. It can easily identify that the target person is about to fall, improve the efficiency of identifying the falling behavior, and avoid the occurrence of falling. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0019] Figure 1 This is a simplified structural diagram of the 3D vision-based human body fall recognition algorithm proposed in the present invention. DETAILED DESCRIPTION
[0020] In order to make the technical means, creative features, objectives and effects of the present invention easy to understand, the present invention is further described below in conjunction with specific embodiments and drawings, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention.
[0021] See also Figure 1 , a human body dumping recognition algorithm based on 3D vision, comprising the following steps:
[0022] Step 1: Collect the moving image of the human body, wherein an optical image sensor, an infrared transmitter, an infrared receiver, and a camera sensor module are used to collect the moving image of the human body, and the human face is recognized to obtain the standing facial orientation of the human body. The camera sensor module includes an RGB image information component and a depth information component, wherein the RGB image information component is an RGB CMOS sensor, and the depth information component is composed of an infrared transmitter head and a CMOS for receiving reflected infrared light. The field of view of the RGB sensor is 58° horizontally, 45° vertically, and 70° diagonally; the spatial resolution at a distance of 2 meters from the sensor is 3mm, and the depth resolution is 1cm; the maximum image acquisition rate is 60fps; the optimal working distance range of the sensor is 1.2~3.5m. ;
[0023] Step 2: Obtain the image information collected in step 1, and detect and separate the image objects in the image information;
[0024] Step 3: Tracking the feature points on the human target model separated in step 2, wherein the collection positions of the feature points on the human target model include the head, neck, chest and legs;
[0025] Step 4: Identify the tendency of the target through the acquired feature points;
[0026] Step 5: Obtain the target inclination direction through the target inclination recognition information, and determine the state of the target. The human body movement state includes the normal upright movement state, the state about to fall, and the falling state. The state about to fall is determined by the angle between the target center of gravity and the ground.
[0027] like Figure 1As shown, a moving image of a target human body is first obtained, and the target human body image in the image is detected and separated from the background, and then the moving trajectory of the target person is tracked, and the target's movements are identified to determine whether the target person is in motion. When the person is in a non-motion state, the target's falling direction is obtained, and then the angle between the falling direction of the target person image and the vertical direction is calculated, so as to facilitate the judgment of whether the target person will fall, thereby improving the efficiency of recognizing the falling behavior and avoiding the occurrence of the falling situation.
[0028] In step 2, the moving target is detected by detecting the changed area, and the first-order image gradient information is obtained by combining the Sobel operator to obtain the target segmentation, and the information after the target human body detected in the image is separated from the background is obtained, which greatly reduces the amount of target detection calculations in the human body recognition process.
[0029] Step 2 also includes distance detection. The distance detection obtains the RGB image through the camera sensor module and inputs it into the SoC sensor processor to obtain the distance between the image target and the camera. The SoC controls the infrared emission CMOS and the depth image CMOS to work simultaneously. The infrared emission CMOS emits coded light, while the depth image CMOS receives the coded light reflected from the scene and inputs it to the SoC in the form of an infrared spectrum. The SoC decodes the coded light using the depth acquisition algorithm based on the information obtained by the visible light image sensor and calculates the depth information of each pixel.
[0030] In step 4, the process of identifying the tendency of the target using the acquired feature points is as follows:
[0031] Step 1: Obtain the position of feature points in space at a certain moment through the two-dimensional information and depth information of the human body:
[0032] Step 1: Obtain the position of feature points in space at a certain moment through the two-dimensional information and depth information of the human body: Step 4: According to the first to third steps, the vector between the legs and the chest can be obtained to determine whether the target is in a moving state, thereby avoiding the error caused by calculating the angle between the straight line where the head and leg feature points are located and the vertical direction during the movement of the target person. When the target person is determined to be in a non-moving state, the angle between the straight line where the head and leg feature points are located and the vertical direction is calculated. The non-moving state here is equivalent to the target person being stationary or suddenly encountering an obstacle and stopping while moving, or the target person being tripped by an object.
[0033] Preferably, the state of being about to fall described in step five includes a state of the target person leaning forward and backward and a state of the target person leaning left and right, wherein the state of the target person leaning forward and backward is when the angle between the target's forward and backward inclination direction and the center of gravity direction is greater than 12.5°; the state of the target person leaning left and right is when the angle between the target's left and right inclination direction and the center of gravity direction is greater than 16°. When the angle between the straight line where the target person's head and chest are located and the vertical direction exceeds a defined value, the target person is in a state of being about to fall, and at this time, the target person can be protected by external factors to prevent the target person from being injured or in danger.
[0034] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A human body tilting recognition algorithm based on 3D vision, characterized in that: The following steps are involved: Step 1: collecting human body moving images, wherein an optical image sensor, an infrared transmitter, an infrared receiver, and a camera sensor module are used to collect human body moving images, and human face recognition is performed to obtain the human body standing face orientation, wherein the camera sensor module includes an RGB image information component and a depth information component; Step 2: Obtain the image information collected in step 1, and detect and separate the image objects in the image information; Step 3: Tracking the feature points on the human target model separated in step 2, wherein the collection positions of the feature points on the human target model include the head, neck, chest and legs; Step 4: Identify the tendency of the target through the acquired feature points; In step 4, the process of identifying the tendency of the target using the acquired feature points is as follows: Step 1: Obtain the position of feature points in space at a certain moment through the two-dimensional information and depth information of the human body: Step 4: According to the first to third steps, the vector between the legs and the chest can be obtained to determine whether the target is in a moving state, and then the angle between the straight line where the head and leg feature points are located and the vertical direction is calculated; Step 5: Obtain the target inclination direction through the target inclination recognition information, and determine the state of the target, where the human body movement state includes normal upright movement state, about to fall state, and falling state. The about to fall state is determined by the angle between the target center of gravity and the ground.
2. The human body falling recognition algorithm based on 3D vision according to claim 1 is characterized in that: In the step 2, the moving target is detected by detecting the changed area, and the first-order image gradient information is obtained by combining the Sobel operator to obtain the target segmentation, and the information after the target human body detected in the image is separated from the background is obtained.
3. The human body falling recognition algorithm based on 3D vision according to claim 1 is characterized in that: The step 2 also includes distance detection, which obtains an RGB image through the camera sensor module and inputs it into the SoC sensor processor to obtain the distance between the image target and the camera.
4. The human body falling recognition algorithm based on 3D vision according to claim 1 is characterized in that: The state of about to fall described in step five includes the state of the target person leaning forward and backward and the state of the target person leaning left and right. The state of the target person leaning forward and backward is that the angle between the target's forward and backward inclination direction and the center of gravity direction is greater than 12.5°; the state of the target person leaning left and right is that the angle between the target's left and right inclination direction and the center of gravity direction is greater than 16°.
