Depth Frames of PrimeSense Camera
The PrimeSense PS1080 camera connects to several computers via an USB 2.0. The IR light projects a patttern of IR dots from the sensor and detects them by a CMOS image sensor with an IR filter. The camera can provide up to 640⨯480 resolutions at a rate 30 frames per second(fps). The OpenNI (natural interaction) provides audio and visual peripherals for PrimeSense PS1080. The camera can give coordinates to the depth of the object. In this system, the position of each point is given by 3 coordinates -- x, y and z [7].
Body height
Humans know how to measure height objects by measuring the angle of sunlight or by using a ruler for a long time.
Save your time!
We can take care of your essay
- Proper editing and formatting
- Free revision, title page, and bibliography
- Flexible prices and money-back guarantee
Recently there have been many projects of height determination using cameras, which is the project of Sotirios Ch. Diamantas and prithviraj Dasgupta, C-MANTIC Lab, University of Nebraska, Omaha, they estimated the height according to the formula:
H_c/h_cp =H_o/h_op
in the formula Hc is the body height of the camera from the ground in cm, h_cp is the body height of the camera in pixels, h_op is the body height of the object -we wish to find its real in pixels, and Ho is the unknown variable, that is the body height of the object in cm, specifically presented in [4].
Since Microsoft released Kinect cmaera in 2012, especially after the release of Microsoft SDKs (Software Devel-opment Kits), it has provided a set of functions such as a skeleton tracker, then many algorithms have developed with applying skeleton points.
N. S. Suriani and colleagues gave a method to identify human actions based on skeleton points [5], projects of this type often serve to develop action game genres.
F. Gossen and T. Margaria offers a method of facial recognition by identifying characteristics between distances of the Kinect's skeleton model, presented in 'Comprehensible people recognition using the Kinect's face and skeleton model,' 2016 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR), Cluj-Napoca, 2016, pp. 1-6. That helps us come up with the idea of calculating the distance between skeleton points to output human height.
Body Weight
There are two basic ways to measure human weight: balance and spring balance. The first way is used to compare mass between objects and the second way to calculate human mass in gravitational environment. So how to calculate human mass in microgravity environment?
K. Shimada and Y. Fujii introduced the tool spring-mas device to do this presented in Key Engineering Materials Vols. 381-382 (2008) pp 461-464.
Carmelo Velardo estimated height by relying on features quan trọng với các phần khác của của của của trường của bạn và đọc được đầy đủ body (height), the torso (waist), and the limbs (arms and legs measures Then use statistical analysis method with a big data of NHANES (National Health and Nutrition Examination Survey) to deduce human mass, presented in [3].
Based on the method of Carmelo Velardo, we provide a method of estimating human mass based on the correlation between height and weight of human through BSA index, the body surface area (BSA), is the measured or calculated surface area of a human body.
Method
Method of determining the height
We choose the tracking body method to estimate the human height. We exploit the skeleton tracker embedded in the OpenNI framework and NITE 2 that provides us with the location of skeleton points, is presented in [1], [2]. It allowed us to follow the human skeleton that was characterized by the skeleton points. Each matching point is represented by its 3D coordinates. There are 15 skeleton points shown in Fig. 1 for our program, with higher versions can give more points.
Fig. 1. The image the skeleton points provided by PrimeSense OpenNI framework [3]
After we have the 3D coordinates of the skeleton points, our next task is to select the appropriate points from head to toe and calculate the total distance between them. Here we select the top down points in the following order: HEAD, NECK, TORSO, LEFT_HIP, LEFT_KNEE, LEFT_FOOT. We can choose the right leg instead of the left leg. The result remains unchanged in the condition that the user stands balanced at the position. The following is the formula used to calculate the total distance between the selected points:
i=26(ψ_(i|x)-ψ_(i-1|x) )^2+ψi|y-ψi-1|y2+ψi|z-ψi-1|z2
in the formula i corresponds to the sequence number of points; x, y, z is the coordinates of each point respectively. The above formula is also equal to the human height.
Method of determining the weight
Through many studies worldwide, scientists have confirmed that there is a correlation between height, weight and body surface area (BSA). Scientists have come up with many formulas that show that correlation.
The most common use is Du Bois formula, 1916:
BSA=0.007184⨯W^0.425⨯H^0.725 (1)
A simpler, easier to use formula is Mosteller, 1987:
BSA=√((W⨯H)/3600)
There are also a number of other recipes including:
Haycock formula, 1978:
BSA=0.024265⨯W^0.5378⨯H^0.3964
Gehan and Gearge formula, 1970:
BSA=0.0235⨯W^0.51456⨯H^0.42246
Fujimoto formula, 1968:
BSA=0.008883⨯W^0.444⨯H^0.663
Takahira formula, 1968:
BSA=0.007241⨯W^0.425⨯H^0.725
Shuter and Aslani, 2000:
BSA=0.00949⨯W^0.441⨯H^0.655
Schlich, 2010:
BSA=0.000975482⨯W^0.46⨯H^1.08 (women)
BSA=0.000579479⨯W^0.38⨯H^1.24 (men).
In the above formulas BSA is in m2, W is weight in kg, and H is height in cm.
Thus, we can see from formula (1) if the value of body surface area (BSA) is determined, then the height measurement above can be calculated from the value of body weight because BSA value fluctuate very small, from 1 to 1.9. We chose the formula because this is the most widely used formula in the world. The formulas is:
W=〖BSA〗^(1/0.425)⨯(H⨯100)^(-0.725/0.425)⨯〖0.007184〗^(-1/0.425) (2).
In this formula BSA is in m2, W is weight in kg, and H is height in cm.
The important thing is how to determine body surface area (BSA) the most appropriate for all subjects. According to research results of a group of Can Tho University of Medicine and Pharmacy students, Vietnam, 2013-2014. Accordingly, the body surface area (BSA) average for men is 1.67 m2 and for women is 1.45 m2. In order to verify these values, we performed a body height and body weight test with 60 people including 30 men and 30 women, the results were close to the above values, specifically the body surface area (BSA) average for men is 1.7 m2 and for women is 1.43 m2. Apply statistical method according to normal distribution with variance σ = 0.069, reliability 1- α = 0.95 or α = 0.05, 〖 Z〗_(α/2)= 1.96 and standard deviation ε = 0.017, then the sample size is:
n=(〖〖 Z〗_(α/2)〗^2⨯σ^2)/〖ε 〗^2 ≈60
thus, the application of statistical probability for such sample size is reliable.
In summary i used the formula (2) with two option the body surface area (BSA) average for men is 1.7 m2 and for women is 1.43 m2 to estimate weight.
Our work
This section will provide an overview of the steps in the calculation method as well as the system that we have implemented. Detail of the sections are presented as shown in Fig. 2. My method is carried out mainly through three steps.
The first is through the process of determining the body height. We use OpenNI framework and NITE 2 [6] to extract skeleton points, then program give us the depth coordinates of the skeleton points in millimeters.
The second, we use my algorithm to calculate the total distance between the selected points as in part II.A to calculate the body height.
The final step is to apply statistical methods to estimate the weight of the object. For this last step, all the estimated estimation tools have been tested.
Fig. 2. Overview of methods of measuring height and weight
Result
The result determines the height
Next, we will present the measured results from the experiment. The purpose of the project is to determine the most appropriate method so that it is possible to accurately estimate the human height and reduce errors. To evaluate the effectiveness of my method use the cumulative square error distribution function:
ε=|λ_m-λ_r |/λ_r (2)
in that, ε is the measurement error, λ_m is the measured value and λ_r is the real value.
Conducting a survey of over 60 people, the average error of measuring is shown in TABLE I. We found that the error of height measurement is ±10% and the average error is 4.37% for men and 4.19% for women, an acceptable value in the medical community.
THE AVERAGE ERROR OF MEASURING
Height Weight
Men 4.37 11.35
Women 4.19 8.8
The result determines the weight
Also from the error estimation method like the formula (2) when measuring the data set above, we have determined the average error of weight for men is 11.35%, the average error of weight for women is 8.8%.
In an experiment conducted at Melbourne Hospital to collect images estimating the volume of 1137 patients and medical staff. The study propose the statistics of the estimate conducted by the patients and the medical personnel. The weight of each patient was firstly estimated by himself, secondly the nurses and the physicians were asked to estimate it. The precision of ±5% the weight of the patients is achieved by the patient and by most of the trained nurses. The physicians instead do not achieve the same results and their error is more spread (up to ±20% of the original weight) [3]. This analysis shows that our system is still effective compared to the estimated weight by visual of medical professionals.
We understand that this analysis is far from practical, not always high-precision compared to other good measuring tools like. Therefore, we studied to evaluate the loss of efficiency and conditions affecting the error of measurement results. We want to prove that regardless of the loss of accuracy in any condition, my measuring system is still considered good enough to use in conditions that do not give you accurate results.
First, we will talk about the condition of the camera device. My program is compatible with Depth Frames of PrimeSense Camera or Kinect v1 Camera. Feature characteristics of the two devices are basically the same. Next, need to set the device so that the camera can take the whole body. Appropriate distance for measurement from body to camera is from 1.2m to 4m, the most suitable distance is 2.5m and minimize obstructions around.
In addition, the brightness factor also affects the measurement, the device will not catch skeleton points in outdoor light conditions, so conducting measurements in the room is most appropriate. Clothing also affects the results of the measurement, loosening the shirt and pants or wearing too thick clothes increases the probability of the measurement error.
My measuring system is not limited to body height measurements. The most accurate measurement limit of real weight measurements is between 50 kg and 65 kg for men and 45 to 60 kg for women, which is common with the physical characteristics of Vietnamese people, outside this value the measurement has an increasing error.
Conclusion
Thus, we have presented a method to estimate the human body height and body weight based on a great skeleton tracking of Kinect feature and a statistical analysis to find the correlation between body height and body weight by Du Bois formula. Models are considered under different conditions by random noise and find conditions for the system to work best. We also demonstrated our system theory better than an estimated experiment based on the human eye. We believe this approach will serve future possible approaches to this topic.
Stuck on your essay?
