Designing a Human Head Circumference

Study Case Designing a Human Head Circumference Measurement System with Machine Vision Michael Kurniawan (2018123456) University of Cincinnati, 2600 Clifton Ave, Cincinnati, OH 45221, United States kurniaml@mail.uc.edu Abstract This study aims to develop an automated anthropometric system based on machine vision that is integrated into a medical cyber-physical system (MCPS) to measure human head circumference. Head circumference is an important parameter in growth monitoring, especially in the detection of disorders such as microcephaly and macrocephaly, which can affect cognitive development and overall health. To address this challenge, this study proposes an automated anthropometry system capable of performing accurate, noncontact measurements that can be accessed in real-time by medical personnel.The system is designed using a machine vision approach, utilizing object detection technology and ellipse model circumference calculations for noninvasive head circumference estimation. A 1920 × 1080 pixel (1080p) camera with a speed of 30 fps and a 60° field of view is mounted on a three-axis mechanism based on a stepper motor to automatically record images of the head from the front and side angles. The measurement stage begins with head object detection and bounding box adjustment to obtain head width parameters. Head circumference estimation is performed using the Euclidean distance method combined with an elliptical geometry approach. Testing shows a minimum error rate of 2.29% at a distance of 50 cm with 300 lux illumination. Evaluation using a confusion matrix resulted in an accuracy of 92.8%, precision of 100%, recall of 97.5%, and F-measure of 98.7%. Keywords: Anthropometry System, Head Circumference, Non-contact Measurement, Machine Vision, Ellipse Perimeter, Medical Cyber-Physical System MICHAEL KURNIAWAN 1 Study Case INTRODUCTION Anthropometry is the discipline that studies the measurement of human body dimensions and proportions. This science is used to understand variations in body size between populations and to detect changes due to environmental factors, lifestyle, or certain medical conditions [1]. There are two main approaches in anthropometry: traditional and modern. Traditional methods use simple tools, such as measuring tapes or calipers, which are inexpensive and easy to use. However, the results are often inconsistent because they depend on the skill of the measurer and are prone to subjective bias. In contrast, modern methods utilize two-dimensional (2D) image processing [2], [3] and threedimensional (3D) modeling [4], [5], for example through 3D scanners, photogrammetry, or a combination of sensors. This approach is capable of producing a more precise and reliable morphological representation of the body than manual methods. As technology develops, the concept of cyber-physical systems (CPS) has expanded to various sectors, including healthcare. CPS is the integration of computing systems and physical entities through digital networks, creating a smart, adaptive, and connected environment. In a medical context, this concept is applied through medical cyberphysical systems (MCPS) [6], which enable fast, accurate, and efficient measurement of body parameters. MCPS is relevant for diagnosis, patient condition monitoring, and rehabilitation, while also supporting the development of personalized and automated precision healthcare services. This study focuses on 2D-based head circumference measurement. Head circumference is an important indicator in monitoring individual growth and development [7]. This measurement is useful for detecting potential disorders such as microcephaly or macrocephaly [8] that can affect cognitive development and general health. Early detection is crucial so that medical intervention can be carried out in a timely manner [9]. In addition, head circumference monitoring helps medical personnel evaluate patient health, especially in conditions that limit direct physical interaction between doctors and patients. The proposed method adopts an image-based telemetry system, designed to be easily accessible by both patients and medical personnel. This technology enables remote measurement without the need for visits to health facilities, making it particularly beneficial for individuals with limited mobility [10]. In addition to supporting long-term MICHAEL KURNIAWAN 2 Study Case monitoring with high accuracy, this approach also reduces the potential for subjective bias from manual methods. Consistent and standardized data enables more objective analysis of head growth and data-based diagnosis. While others utilize OpenPose to detect body parts such as chest, waist, and hip circumference, then calculate them using the ellipse circumference formula [1]. Another study applied an accelerometer and gyroscope-based telemetry system to detect human falls through a wearable device [11]. A similar approach was also used in biosignal measurement [12], although it required direct contact with the user's body. Other methods examine body dimension measurements in 2D and 3D with convolutional body dimensions, point cloud, skinned multi-person linear (SMPL), and non-linear regressor [13]. There is also research on body weight estimation using the body surface area approach through the ellipse tube formula and linear regression [14], [15], as well as the Viola-Jones method for object detection in waist circumference measurement [16]. This study proposes an object detection-based telemetry system combined with ellipse circumference calculation to automatically measure head circumference. The camera is mounted on a two-axis mechanism with automatic height adjustment, allowing the system to identify objects independently. The main contribution of this research is the development of an automatic anthropometry system through the integration of MCPS, a multi-axis moving camera, and an ellipse-based object detection algorithm. This approach enables non-contact, real-time, and precise measurements, while overcoming the limitations of traditional methods that rely on manual tools or wearable devices with direct contact [3], [11], [12]. Modern 3Dbased methods, such as SMPL or point cloud [13], are indeed precise, but require expensive equipment and complex setup. The proposed system offers an automated, efficient, and remotely accessible measurement solution via machine vision-based telemetry. This approach is a noninvasive alternative compatible with telemedicine platforms, with the potential to transform growth monitoring. Its advantages enable faster, safer, and more accurate screening, with benefits not only for diagnosis but also for clinical evaluation, nutritional status monitoring, and the design of head protection devices based on valid anthropometric data. The structure of this article is divided as follows: Sections II and III describe the system design and algorithms used, including testing procedures. Section IV presents the experimental results and analysis, while Section V contains the conclusions. MICHAEL KURNIAWAN 3 Study Case SYSTEM PERFORMANCE Overview Figure 1 shows the performance of the telemetry system for measuring human head circumference based on machine vision within the framework of a medical cyberphysical system (MCPS). In the initial block diagram, the MCPS runs the main subprocess of data acquisition through a camera as the main sensor. The captured images are then analyzed to determine the centroid or geometric center of the object. Determining the centroid serves as a reference for stabilizing the position of the object so that the measurement results are more accurate and consistent [17]-[19]. After the centroid is determined, the image is analyzed using a machine vision algorithm for identification and classification based on the morphological characteristics of the human body [11], [12]. The role of machine vision is very important because it enables automatic detection without manual interaction, thereby reducing the potential for human error. In this system, the camera is positioned as the main visual sensor in the MCPS. The device records images of objects, which are then processed automatically, enabling monitoring without direct contact. This is particularly relevant for patients with limited mobility or in epidemic situations, as the system reduces the need for physical interaction. The use of centroids has a primary purpose, namely to determine the average position of objects so that measurements can be stabilized and adjusted [17]. The process of converting sizes from pixels to millimeters is carried out by considering the distance of the camera to the object, which is very important so that the measurement results correspond to the actual dimensions. In addition, this study also corrects perspective and geometric distortion due to different shooting angles. To that end, a recalibration process is applied to improve accuracy. In determining the centroid, the image moment method is used, which allows the calculation of the radius, area, and center of mass to determine the center coordinates (x, y) of the object contour [18]. The system also integrates an object detection algorithm to recognize humans and other objects in the vicinity. Object detection is classified into two categories: soft detection, which only identifies the presence of objects without detailed location, and hard detection, which determines the location and size of objects with precision [20]. This stage forms the basis for further anthropometric analysis. MICHAEL KURNIAWAN 4 Study Case Anthropometry in this study covers two categories: static, which focuses on body shape and composition, and dynamic, which relates to movement, strength, and use of space in activities [13], [21], [22]. Device specifications and system requirements standards are presented in Table I as a reference for optimal head circumference telemetry measurements. The measurement procedure was carried out with the subject sitting on a chair facing the device to maintain image stability. The system used three camera distance variations (50 cm to 70 cm) to optimize detection and adjust lighting conditions. These distance variations helped determine the optimal position for measurement accuracy. Figure 2 illustrates the image acquisition stages with adjustments to the subject's position and camera angle according to the standards. Meanwhile, Figure 3 shows the horizontal trajectory of the camera, which can move automatically to the left and right using a stepper motor. This movement allows the system to adjust the viewing angle optimally before capturing images for feature extraction [23], [24]. Feature extraction aims to isolate important information, such as the shape and dimensions of the head, to improve the accuracy of the measurement results. With automatic movement on the horizontal path, the system reduces human intervention, thereby minimizing the potential for manual errors. Mechanically, the system is designed with camera mobility paths on two axes: horizontal and vertical. The horizontal path forms an L-shaped pattern with a quartercircle curve, allowing images to be captured from the front and side. The vertical path adjusts the camera position to the height of the object, so that the image is always in the optimal field of view. The combination of these two paths makes the system more flexible, efficient, and precise. MICHAEL KURNIAWAN 5 Study Case Figure 1. Performance of the machine vision-based head circumference telemetry system Figure 2. Automatic camera trajectory for image acquisition Figure 3. Horizontal camera trajectory in the image capture and extraction process Table I. Standard specifications of measurement support devices MICHAEL KURNIAWAN 6 Study Case Calculation of Head Length and Width The measurement process begins with the detection of the head object from the front and side angles. A bounding box is used to determine the length (Figure 4a) and width (Figure 4b) of the head. Once the object is detected, the system adjusts the bounding box to measure the width of the head on the horizontal axis (x-axis), as shown in Figures 5a and 5b. The measurement data is then used in an ellipse model to estimate the head circumference. This approach is more accurate because it adjusts to the natural contours of the human head. The estimation results are visualized in Figure 5c as the final output. Head width measurement is performed using the Euclidean distance method, which is the calculation of the distance between two points in coordinate space [13], [18]. The distance value in pixels is then converted to centimeters (cm) using equation (1): Equation (1) is used to calculate the ratio between the object width in pixels (PixelWidth) and the actual head width (AreaWidth). This ratio allows the system to convert pixel-based digital measurements into metric units, ensuring that the measurement results comply with medical anthropometric standards. MICHAEL KURNIAWAN 7 Study Case Figure 4. Head bounding box: (a) front view, (b) side view Figure 5. Ellipse-based head circumference estimation: (a) head width, (b) head length, (c) head circumference estimation results With this approach, the system can produce more precise, consistent, and reliable head circumference measurements for medical applications and anthropometric analysis in human growth monitoring. Head Circumference Calculation At this stage, head circumference is calculated using the ellipse perimeter formula [1], [12], [25], as visualized in Figure 5(c). The results of measuring the length and width of the head from the front and side views are used in the ellipse model as shown in (2). In this equation, a and b represent the semi-major axis and semi-minor axis of the ellipse estimated based on the system's detection results. As validation, manual measurements were taken using a measuring tape as a conventional method. The actual results showed a head circumference of 55 cm, with a front width dimension of 16.4 cm (semi-axis 8.2 cm) and a side width dimension of 17.5 cm (semi-axis 8.75 cm). Calculations using the ellipse model produced estimates that were very close to the manual values, with a difference of about 2 cm. To improve accuracy, this error was MICHAEL KURNIAWAN 8 Study Case compensated for through pixel-to-metric conversion calibration in the algorithm. This adjustment made the system more precise and reliable as a technology-based anthropometric analysis tool. Head Circumference Measurement System Testing Testing was conducted by comparing the system's results with manual measurements using a measuring tape as a reference. The aim was to evaluate the system's accuracy in estimating head dimensions. After the data was obtained, the accuracy level was calculated through error percentage analysis, which showed the deviation of the system's results from manual measurements. The accuracy value is calculated by subtracting the error from 100%. This means that the smaller the error, the higher the accuracy of the system [14]. Equation (3) is used to calculate the error and accuracy: With this calculation, the reliability of the system can be determined. If the accuracy is close to 100%, then the system has good performance and is suitable for use as an alternative method of machine vision-based telemetry measurement. SYSTEM TESTING Figure 6(a) shows the implementation of the system when the object is sitting facing the camera at a calibrated distance. The camera automatically adjusts its position for the optimal measurement angle before detection begins. Figure 6. System implementation: (a) application of telemetry to human objects, (b) bounding box results in the telemetry system. MICHAEL KURNIAWAN 9 Study Case This system integrates machine vision with object detection algorithms to recognize humans and other objects. There are two types of detection: soft detection, which only identifies the presence of an object, and hard detection, which determines the position, size, and boundaries of an object with precision [20]. To ensure that the head meets the measurement criteria, the camera automatically places the object in the center of the field of view with a frontal orientation. The initial stage is frame alignment, which is image alignment using a bounding box. If the object is outside the optimal zone, the camera adjusts its position via an actuator. Additional validation is performed based on facial symmetry, image quality, and visual disturbances such as accessories. Thus, only optimal images are used in the measurement. After successful front and side detection, the system measures the head width as the main parameter for head circumference estimation. This process is visualized in Figure 6(b). The results show that the ellipse approach provides more stable and accurate measurements than the manual method. Front and side image data are stored on the server as a dataset for model training and testing. The initial dataset consists of 420 samples, with an 80:20 ratio for training and testing, so that the model obtains sufficient data and is tested on new samples. Evaluation is performed on front and side images by comparing reference annotations with system predictions. Performance was analyzed using a confusion matrix, which was the basis for calculating accuracy, precision, recall, and F1-score. Additional testing involved humans wearing accessories such as glasses, masks, and hats (Figure 7a). The model still detected heads with black hats because they resembled hair, but had difficulty with faces wearing masks because the dataset did not include such variations. In addition to humans, testing was also conducted on non-human objects such as dolls and animations. The model did not detect these objects because the dataset did not have non-human annotations. However, in some cases, false positives occurred, where nonhuman objects were mistakenly detected as humans (Figure 7b). This was due to visual similarities with humans, especially in terms of facial features and body proportions. MICHAEL KURNIAWAN 10 Study Case Figure 7. System testing: (a) human objects with accessories, (b) false positives on non-human objects. From the test results, it can be concluded that the model detects humans well and is able to distinguish between trained and untrained object classes. However, increased robustness is still needed to overcome variations in facial accessories and reduce false positives on human-like objects. Object Detection Testing Table II shows the test results for several parameters that affect the performance of the object detection algorithm. The parameters tested include variations in light intensity and the distance between the object and the camera. The tests were conducted indoors with three different lighting levels and three variations in the distance between the object and the camera to evaluate the influence of these factors on detection performance. TABLE II. Light Intensity Parameter Testing MICHAEL KURNIAWAN 11 Study Case Based on the test results, the object detection model was able to detect objects well in the light intensity range of 200–400 lux. If the light intensity is less than 200 lux, the system has difficulty detecting due to the low contrast between the object and the background, so that the bounding box is not formed properly. Conversely, at intensities above 400 lux, detection still works, but excessive light reflection sometimes reduces accuracy. The distance of the object from the camera also has a proven effect. Optimal detection occurs at a distance of 50–60 cm. If it is too close (< 50 cm) or too far (> 60 cm), accuracy decreases due to perspective distortion that affects the dimensions of the object in the image. These findings confirm that lighting calibration and determining the optimal distance are important for the system to work optimally. Head Circumference Measurement Testing In testing the head circumference measurement system, the accuracy of the measurement results from the designed system was evaluated, as shown in Table III. The testing involved six object samples, each measured at three different distances to analyze the effect of distance variation on accuracy. During the test, the light intensity was maintained at a range of 300–400 lux, with the camera distance set at 50 cm, 60 cm, and 70 cm. MICHAEL KURNIAWAN 12 Study Case The results from the system were then compared with manual measurements using a measuring tape to calculate the accuracy and error of the system. Based on Table III, the average error of all measurements at the three distances was 2.999%. Of the three, the 50 cm distance provided the best results, with an average error of 2.299%, indicating that the system was most accurate at that distance. Factors contributing to the error include the distance of the object to the camera, the position of the object's body when sitting, and the lighting in the room. These factors affect the detection of the bounding box and the calculation of the head dimensions, so object position stability and optimal lighting are required for the system to achieve maximum accuracy. TABLE III. Head Circumference Measurement Testing Head Circumference Testing of Objects of Different Ages In addition to distance variations, head circumference measurements were also tested based on differences in the age of the objects. The purpose of this test was to ensure that the system could perform accurate measurements across a range of ages, especially children, whose heads are smaller than those of adults. The measurement data based on age parameters is presented in Table IV. The test was conducted on three samples with different age categories: ≤ 10 years, 12 ≤ x ≤ 23 years, and ≥ 24 years. All subjects were measured at a distance of 50 cm to ensure that the results were not affected by distance variations. MICHAEL KURNIAWAN 13 Study Case Based on Table IV, the average measurement error was 2.002%. In the ≥ 24 years age category, the smallest error was 0.179%, indicating that head circumference detection was very accurate in adults. In the ≤ 10 years age category, the error was 1.569%, which was still within reasonable limits. However, in the 12 ≤ x ≤ 23 years age group, the error reached 4.259%, the highest value among the three. These results indicate that the system is more accurate on objects with larger head sizes (≥ 24 years) than on adolescents, who tend to have more complex head shapes. Lighting and object position also affect the results. Therefore, the system can still be optimized to improve accuracy, especially on objects with small or varying head sizes. TABLE IV. Head Circumference Testing Based on Age RESULTS After conducting a series of tests, the results are shown in Table V, which contains the system's performance in detecting front and side objects. This test evaluates the model's success rate based on the confidence score: ≥ 0.9 (highly accurate detection) and < 0.9 (less accurate detection). The results show that most objects were detected well, although some samples had lower confidence values. Factors that influenced this included: 1. Non-human objects: such as dolls or animations were not significantly detected, because the training dataset only included humans. 2. Accessories: the presence of masks, glasses, or long hair covering the face lowered the confidence score. 3. Lighting and position: suboptimal lighting conditions and head positions that differed from the training pattern reduced accuracy. Detailed results are shown in Table VI, which lists the main factors causing a decrease in system accuracy. TABLE V. Front and Side View Dataset Test Results MICHAEL KURNIAWAN 14 Study Case TABLE VI. Front and Side View Dataset Confusion Matrix TABLE VII. Front and Side View Real-time Test Confusion Matrix Based on Table VI, performance metrics were calculated to evaluate the system. The metrics used include Accuracy, Precision, Recall, and F-measure. Calculation results: • Accuracy = 92.8%, indicating the overall accuracy of detection. • Precision = 100%, all detected objects are correct according to the target class, with no errors on non-target objects. • Recall = 97.5%, most objects that should be detected are successfully recognized, although some are not due to lighting or accessories. • F-measure = 98.7%, the harmonic mean between Precision and Recall, indicating balanced system performance. These results prove that the machine vision-based head circumference measurement system has excellent performance with low error. Although lighting, position, and accessories remain challenges, the system demonstrates high accuracy and is suitable as an alternative method for automatic measurement. In real-time testing (Table VII), the following results were obtained: • True Positive (TP) = 20 • False Positive (FP) = 0 • • False Negative (FN) = 0 True Negative (TN) = 0 (not relevant because there were no negative samples) MICHAEL KURNIAWAN 15 Study Case Thus, the system achieved 100% accuracy in real-time conditions, confirming its reliability for human object detection according to the training dataset. CONCLUSION This study concludes that a machine vision-based head circumference telemetry system that integrates object detection algorithms with an elliptical approach for circumference calculation is capable of automating the detection and measurement process with high precision. The relative error rate was in the range of 2.299%–3.494%. This system improves the efficiency and consistency of anthropometric data acquisition and supports the development of biometric technology that can be integrated with smart devices and medical telemetry platforms. Its applications extend from the health sector, the wearable device industry, to real-time population monitoring. However, the system still needs to be optimized, especially for individuals with extreme head dimensions, such as children under 10 years of age. Future development directions include: • improved generalization of morphological diversity, • validation in various environmental conditions, • improved performance in dynamic operational contexts, • application of adaptive mechanisms to maintain consistency in measuring object positions and distances. MICHAEL KURNIAWAN 16 Study Case REFERENCE MICHAEL KURNIAWAN 17