Geri Dön

İnsan hareketi izleme teknolojilerinin karşılaştırmalı değerlendirmesi

Comparative assessment of human motion monitoring technologies

PDF İndir

  1. Tez No: 559423
  2. Yazar: CEMAL FATİH KUYUCU
  3. Danışmanlar: DR. ÖĞR. ÜYESİ GÖKHAN İNCE
  4. Tez Türü: Yüksek Lisans
  5. Konular: Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Computer Engineering and Computer Science and Control
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2019
  8. Dil: Türkçe
  9. Üniversite: İstanbul Teknik Üniversitesi
  10. Enstitü: Fen Bilimleri Enstitüsü
  11. Ana Bilim Dalı: Bilgisayar Mühendisliği Ana Bilim Dalı
  12. Bilim Dalı: Bilgisayar Mühendisliği Bilim Dalı
  13. Sayfa Sayısı: 81

Özet

İnsan hareketi izleme teknolojileri denildiğinde bir bireyin vücudunda yer alan uzuvların oryantosyonu ve pozisyonunun çıkarımı anlaşılmaktadır. Bu uzuvlara yerleştirilen veya harici olarak deneğin bulunduğu ortama konulan sistemler ile hareketin takip edilmesi ve tespiti yapılabilmektedir. Her sistemin kendi içerisinde farklı avantajlar ve dezavantajlar yer almaktadır. Bu çalışma içerisinde farklı özelliklere sahip dört hareket yakalama sisteminin güvenirlilik, kullanım kolaylığı, genellik, maliyet, ortam şartlarına duyarlılık ve hareket esnekliği yönlerinden kıyaslanması amaçlanmıştır. RGB kamera tabanlı, derinlik sensörü tabanlı, atalet ölçüm birimi tabanlı ve giyilebilir tekstil tabanlı sistemler seçilerek, bu sistemlerin incelemenmesine karar verilmiştir. Atalet ölçüm birimi tabanlı sistem tüm vücuda giyilen küçük ataletsel ölçüm birimlerinden ve bunların toplandığı bir merkezi işleyiciden oluşmaktadır. Bu sistemlerin bir araya gelmesiyle birlikte sensör füzyon yöntemleri kullanılarak hareket elde edilmektedir. Derinlik kamerası ile üretilen hareket yakalama sistemleri standart kameralara eklenmiş derinlik algılayıcıları ile çalışmakta ve ortamı analiz ederek tespit işlemini gerçekleştirmektedir. Standart kamera ile yapılan çalışmada makine öğrenmesi yöntemleri ile hazırlanmış uzuv tanıma modelleri kullanılarak insan duruşu çıkarılması sağlanmaktadır. Bu çalışma içerisinde kullanılmış son yöntem ise tekstil tabanlı giyilebilir teknolojidir. Elastik iletken örgü iki kumaş katmanı arasına yerleştirilen silikon katman sayesinde üretilmiş sensörler uzayıp kısaldığında lineer yapıda kapasitans değeri üretmektedir. Bu kapasitans değerinin değişimine dayanarak bir diz ölçüm sistemi geliştirilmiştir. Bu sistem ile diğer sistemlerin karşılaştırılması çalışma içerisinde yapılmıştır. Bu karşılaştırmanın yapılabilmesi amacıyla ataletsel ölçüm tabanlı sistem olarak Perception Neuron, derinlik sensörü tabanlı sistem olarak Microsoft Kinect ve standart kamera tabanlı sistem olarak Open-Pose seçilmiştir. Karşılaştırma alanında yapılacak deneylerde bir doğrulama düzeneğine ihtiyaç duyulmaktadır. Bu doğrulama düzeneği bu çalışma kapsamında analog açı sensörleri ile gerçeklenmiştir. Deney için gerekli olan son bileşen ise görselleştirmenin yapılacağı test arayüzüdür. Unity ortamında yapılan test arayüzü içerisinde tüm sistemlerin ve doğrulama düzeneğinin verilerinin eş zamanlı olarak görselleştirilebilmesi ve deneğin gerçek duruşunun kamera ile izlenmesi sağlanmıştır. Bu arayüz aynı zamanda deney esnasında elde edilen verilerin de kayıt altına alınmasını sağlamaktadır. Normal ortam şartlarında, loş ışıklı ortamda ve farklı duruş açısında yapılan deneylerde yalnızca sağ dizin açılarının hesaplanması için çalışma yapılmıştır. Her deney senaryosu 10 kez dizin kapatılıp açılmasına dayanmaktadır ve her deney farklı özelliklere sahip 5 farklı denek tarafından tekrarlanmıştır. Deneyin tamamlanmasının ardından sonuçlar çalışma içerisinde verilerek çalışma tamamlanmıştır. Sonuç olarak sistemlerin farklı kriterlerde başarılı olduğu ve ihtiyaca yönelik değerlendirmenin faydalı olacağı tespit edilmiş olmakla birlikte giyilebilir tekstil tabanlı sistemin tüm kriterler için başarılı sonuçlar verdiği görülmüştür. Gelecek çalışmalarda bu karşılaştırmanın diğer eklemler için de genişletilmesi ve denek sayısının artırılarak daha fazla senaryonun denenmesi planlanmaktadır.

Özet (Çeviri)

The use of humanoid models in computer environments is necessary for many different fields such as simulation, gaming, films, animations, and architectural design to name a few. Various computer programs are used in this field for the preparation of models. It is up to the user to capture the details in the modeling process. Thus, even a small animation can take very long to prepare. Various methods have been introduced in order to overcome these issues and to capture the motion characteristics of human-like models. These methods are called motion capture (mocap) systems. When discussing human motion monitoring technologies, we mean the extraction of the orientation and position of the limbs of an individual. The motion tracking information can be obtained by means of sensors placed on the limbs of the individual or systems which exist in and monitor the environment of the individual. Both systems have advantages and disadvantages. This study aims to compare and contrast four motion capture systems of varying characteristics. The four motion capture methods chosen for comparison are an inertial measurement unit (IMU) based system, a depth sensor system, a standard camera and a textile sensor based system. The IMU based mocap system consists of a system of small inertial measurement units worn around the entire body and a central unit where the collected data is accumulated. This data is then processed in the computer. The group of sensors and a method of sensor fusion is used to obtain motion characteristics. In the second method, machine learning methods which implement limb detection models are used together with the standard camera feedback to extract human posture. The third technology compared in this study is the depth sensor. These sensors measure the distance of points in their field of view by transmitting invisible infrared lights and measuring the time of flight after the light reflects off objects. As depth sensors were introduced to the software development environment, studies which utilized them for posture detection have been conducted increasingly. The last method of motion capture used in this study is a textile sensor based wearable technology. Sensors produced by placing a silicon layer between two elastic conductive cloth layers produce linear capacitance values when stretched and released. For the textile sensor based system, the capacitance values obtained from the elastic sensor are read by the analog measuring unit and transferred wirelessly to a processor where they are converted to angle values a visual is created based on these values. The wearable system consisting of sensors and a central processor is able to share data with other devices using Bluetooth technology. Elastic strips cut from this specially produced fabric are placed on the user's joints. The strips stretch and contract with the movement of bones at both ends of the joint. The system is calibrated after collecting highest and lowest tension readings. Then, stretch values of the textile sensor based system are converted to angle values and a model is created. The comparison of the four of these systems has been done in the scope of this study. In doing these experiments, the Perception Neuron system was chosen as the IMU based mocap system, Microsoft Kinect was preferred as the depth sensor based mocap, and Open-Pose was decided on as the standard camera based mocap. A validation mechanism is necessary so that the experiments can be compared. The validation mechanism consists of an analog angle sensor which is mounted on the coupling axis of the two rods which can be worn on the user's knee. For the purposes of this study, the Me Angular sensor by Makeblock was preferred because it produces infinite rotation analog output. Using the mechanism prepared with the use of this sensor, measurements were made. These measurements were passed through an average value filter in order to reduce noise and transferred wirelessly to the unity environment to be used as reference values. The final component necessary to conduct experiments was a test interface where real-time visualization of the experiment components and errors will be made possible. This interface contains models created with the real avatar image taken from the camera and the data obtained from other experimental components. The humanoid system for each experiment component is selected and positioned in the same way. Real-time data and errors are shown along with the corresponding avatar for each experiment component. In the test interface developed using the Unity environment, it is ensured that the data coming from all four systems along with the measurements of the verification mechanism can be seen simultaneously. This platform allows transferring motion information to an avatar in many ways. Because joint tension data is collected from the textile sensor based sensor, joint motion modeling is utilized for the purposes of this study. The actual pose of the subject monitored by the camera is also presented in this interface. This interface also allows the recording of the data obtained during the experiment for analysis at a later time. In the experiments, measurement error can be calculated quantitatively. A reliable system must have a high measuring accuracy. For this reason, an error is obtained for the success of each system by averaging the errors between the reference system and other systems. In order to evaluate success, measurements in different ambient conditions and reliability in these conditions were evaluated separately. Another key criterion is the ease of use. The preparation steps before the test represent the ease of use of the system. The amount of time spent in the preparation phase demonstrates the success/failure of the systems in terms of ease of use. Portability is yet one more criterion to consider. The capability to use these systems in different environments is a significant advantage for the system. In this experiment, all systems have been mounted in the most portable way. During the evaluation, the power and data communication method of the system has been taken into consideration. The last assessable success criterion is the concept of generality. Whether the existing measurement systems can be used easily by other users determines the generality of this system. Consequently, reliability, ease of use, cost, and sensitivity to environmental conditions will be evaluated quantitatively, and generality and flexibility of movement will be evaluated qualitatively. Tests were carried out on five different subjects. The height of these subjects ranged from 1.7 meters to 1.85 meters and the weight of the subjects ranged from 70 kg to 130 kg. The group consisted of male and female subjects, all students in ITU, aged between 22 and 30 years old. In this study, the right knee movement measurements were taken from subjects in a sitting condition. This was done so that the experimentation phase was easy on the subject and so that the motion capture systems could be compared under equal conditions. A seating arrangement where the subject can conveniently be seen by both camera-based systems at an angle of 45 degrees and a distance of 2 meters. As each experiment was conducted, the data was collected by all four systems simultaneously. Two different light elements were tested in the experimental environment. Scenarios where ambient lighting is high and low are examined in the study. In order to investigate the effects of the camera positioning angles on the performance of the systems, the positions of the subjects in front of the camera were also changed with different repetitions of the experiment. All systems were tested simultaneously for consistency of experiments. The testing environment was first prepared before the experiment. The chair position is centered within the field of view of the cameras. The two wearable systems that need calibration are calibrated. After the calibration process of the mocap systems, the previously calibrated validation mechanism is fixed to the subject's legs. The final procedure before the experiment is the positioning of the depth sensor and standard camera. In order to evaluate ease of use, the time it takes to set up, wear or calibrate systems has been recorded. To conduct the experiment, the subject is required to bend his knee to 90 degrees (closed position) and then stretch his knee out to 180 degrees (open position) to determine the optimal performance of all systems. This process is repeated 10 times and measurements are taken. Afterward, the experiment is repeated in a dim environment by reducing the amount of light to evaluate the ambient conditions. Similarly, the subject starts with his right foot in the closed position and repeats the open-close movement 10 times. Next, the lighting conditions are brought back to normal and the angle of the subject with respect to the cameras is changed. The experiment is then repeated. The data obtained during all test steps were recorded in a CSV file so that it could be processed at a later time. The aim of this thesis is to compare the methods used to model human body movements in digital environments. In the scope of this study, the preliminary investigations were made for the comparison process. As a result of the research conducted on each system, a set of comparison criteria has been determined by analyzing situations where the systems were successful and unsuccessful. Finally, the necessary libraries and working environments necessary for all systems have been prepared and experiments have been conducted. In order to visualize the systems simultaneously, a 3D stage has been prepared. In addition to the technologies being compared, a validation mechanism has been prepared and calibrated to compare the success of the mocap systems to a ground truth value. When the results of the experiments are examined, it can see that the textile sensor based wearable system is successful in terms of reliability. In general, the on-body mocap systems produce results similar to each other and better than the camera based mocap systems because the camera based systems are easily affected by ambient conditions. Another criterion determined at the beginning of the study was ease of use. The camera based systems are the most successful when it comes to ease of use because they do not generally have difficult setup and calibration steps. According to the results of the experiment, the IMU based mocap system is harder to wear, therefore falls behind the textile sensor based wearable mocap system when it comes to ease of use. While the IMU based mocap system can be considered successful in many comparison areas, this technology is not as accessible due to its high cost. The RGB camera-based system will be able to perform acceptably at low costs. Moreover, using multiple devices may be able to achieve a success rate similar to the IMU based mocap system. The textile sensor based wearable system can be obtained at a low cost which was one of the objectives in its production. Generality was also listed in the set of evaluation criteria. In this sense, the IMU based system is one of the most difficult systems to use due to the need for limb measurements for each user. The standard camera based system also doesn't meet the generality condition because it only recognizes specific types of outfit. The textile sensor based wearable system comes after the standard camera based system because it also depends on the body measurements of the user. Thus, the depth sensor based system can be considered the most successful in terms of the generality criterion because it is not dependent on the individual using it. In terms of flexibility of motion, the camera based systems fall behind because they are limited to a certain field of view. Because the IMU based system and the textile sensor based wearable system require portable batteries they are more preferable. When the results of the experiments and research are examined, it is seen that all four systems are functional when it comes to executing the basic task of motion capture. The success ranking of each system varies according to different criteria. This shows that systems can be selected according to the project needs. Because these end-user products are generally intended for amateur use consumers are more concerned with the cost of the product. Taking this into consideration and evaluating all the systems overall, it can be said that the wearable textile sensor based system is a success. In future studies, it is planned to expand this comparison for other joints, to increase the number of subjects, and to try varying scenarios.

Benzer Tezler

  1. Tez No

    421066

    Natural ventilation of high-rise buildings a methodology for planning with different analysis tools and case-study integration

    Çok katlı binalarda doğal havalandırma farklı analiz araçları ve örnek alan entegrasyonu ile planlama için bir yöntem

    TOBIAS SCHULZE

    Doktora

    İngilizce

    İngilizce

    2015

    Enerjiİstanbul Teknik Üniversitesi

    Mimarlık Ana Bilim Dalı

    Prof. Dr. AYŞE ZERRİN YILMAZ

    PROF. DR. MARCO PERINO

  2. Tez No

    847190

    Prediction of COVID 19 disease using chest X-ray images based on deep learning

    Derin öğrenmeye dayalı göğüs röntgen görüntüleri kullanarak COVID 19 hastalığının tahmini

    ISMAEL ABDULLAH MOHAMMED AL-RAWE

    Yüksek Lisans

    İngilizce

    İngilizce

    2024

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolGazi Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    PROF. DR. ADEM TEKEREK

  3. Tez No

    467256

    Funport: An interactive public display that enhance the waiting experience at airport boarding lounges

    Havalimanlarındaki bekleme deneyimini zenginleştirmek için etkileşimli kamusal ekran önerisi: Funport

    CANDAN ERİM

    Yüksek Lisans

    İngilizce

    İngilizce

    2017

    Mimarlıkİstanbul Teknik Üniversitesi

    Bilişim Ana Bilim Dalı

    DOÇ. DR. YÜKSEL DEMİR

  4. Tez No

    866262

    Enhancement of situational awareness in physical security using mixed reality

    Fiziksel güvenlikte karma gerçeklikle durumsal farkındalığın artırılması

    NAZIM YİĞİT KAVASOĞLU

    Yüksek Lisans

    İngilizce

    İngilizce

    2023

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    DOÇ. DR. GÖKHAN İNCE

  5. Tez No

    564939

    Creation of web-based application for earthquake hazard analysis by GIS and AHP

    CBS ve AHP yardımıyla deprem tehlike analizi için web tabanlı bir uygulamanın oluşturulması

    IBRAHIM S.M. FARROUKH IBRAHIM S. M. FARROUKH

    Yüksek Lisans

    İngilizce

    İngilizce

    2019

    Coğrafyaİstanbul Teknik Üniversitesi

    Bilişim Uygulamaları Ana Bilim Dalı

    DOÇ. DR. TURAN ERDEN

Geri Dön