M.S. Theses

Permanent URI for this collection

https://digitalarchive.library.bogazici.edu.tr/handle/123456789/11669

Browse

Now showing 1 - 20 of 74

Modular safety-critical control of legged robots
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2022., 2022) Tosun, Berk.; Samur, Evren.
With recent performance improvements, legged robots will soon enter our lives to stay. Various control algorithms are already employed in deploying existing robots, and many more algorithms are still in the making. Safety concerns during the operation of legged robots must be addressed to enable their widespread use and ease their development. Especially machine learning- based control methods would benefit from model-based constraints to improve their safety. This thesis presents a modular safety filter to improve safety, i.e., reduce the chance of a fall of a legged robot. The availability of a robot capable of locomotion is assumed, i.e., a nominal controller exists. During locomotion, terrain properties around the robot are estimated through machine learning which uses a minimal set of proprioceptive signals. A novel deep-learning model utilizing an efficient transformer architecture is used for terrain estimation. A quadratic program combines the terrain estimations with inverse dynamics and a novel control barrier function constraint to filter and certify nominal control signals. The result is an optimal controller that acts as a filter and the filtered control signal allows the safe locomotion of the robot. The resulting approach is general and could be transferred with low effort to any other legged system.
A reinforcement learning based controller to minimize forces on the crutches of a lower-limb exoskeleton
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2022., 2022) Utku, Aydın Emre.; Samur, Evren.; Öncü, Sinan.
The majority of the metabolic energy consumption of a lower-limb exoskeleton user comes from the upper body effort, since the lower body can be considered to be passive. However, the upper body effort of lower limb exoskeleton users is ignored during motion controller development process in the literature. In this thesis study, deep reinforcement learning is used to develop a locomotion controller that minimizes the ground reaction forces (GRF) on crutches. The rationale for minimizing the ground reaction forces is to minimize the upper body effort of the user. A model of the human-exoskeleton system with crutches is created in URDF and XML formats. Reward functions that encourage the forward displacement of the center of mass of the exoskeleton-human system without falling and extreme joint torques are shaped. The state-of-the-art methods, Twin Delayed Deep Deterministic Policy Gradient (TD3) and Proximal Policy Optimization (PPO), are employed with the RaiSim and MuJoCo physics simulators and with different algorithm specific parameters in multiple training trials. The employed networks generate the joint torques based on the joint angle and velocities along with the ground reaction forces on feet and crutch tips. These generated joint torques are directly sent to the exoskeleton model and a new state is observed after implementing the action that the deep RL framework provides. Policies trained by the TD3 and PPO methods on RaiSim are observed to fail to generate proper control commands for a stable and natural looking gait. In general, it is observed that the PPO method generated higher rewards than the TD3 method on RaiSim. After failing to develop a desired policy with RaiSim, MuJoCo is employed as the simulator. Eventually, a policy that can generate a reasonable gait with a desired crutch usage and with 35% minimization in GRFs with respect to the baseline policy is developed.
Design of a social robot and safe social navigation with deep reinforcement learning
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2022., 2022) Bektaş, Kemal.; Bozma, H. Işıl.
This thesis is concerned with the design and development of a social robot that can navigate around in a socially compliant manner. The importance of this problem is due to the growing demand of using robots in human-populated environments. In this thesis, this problem is addressed in two concurrent parts. The first part has focused on the physical design and development of a social robot - named as SempRob. SempRob is aimed to have a sympathetic appearance while also having a design in which its visual sensors are located appropriately for environmental sensing. In the second part, the social navigation capability of the social robot is developed. First, a novel navigation method referred to as artificial potential function with reinforcement learning (APF-RL) method. In addition, an ellipse-based representation of obstacles is developed for efficient obstacle representation. Furthermore, environmental complexity measures are defined in order to ensure that learning scenarios incorporate a range of maneuvering difficulties. Both simulation and experimental results with SempRob demonstrate that APF-RL method enables the robot to move safely and efficiently in complex environments. Following, APF-RL method is extended to Social APF-RL method so that the robot additionally respects the comfort zones of the humans while navigating. This requires the robot to detect the humans in its surroundings and to track them spatially. A deep learning based human detection algorithm is combined with a Kalman filter for this purpose. Finally, Social APF-RL method is modified to be applicable in human following as well. All the proposed methods are tested on the developed robot successfully.
Personalized product recommendation on second hand platforms
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2022., 2022) Yarar, Ramazan.; Baydoğan, Mustafa Gökçe.
With the advent of online marketplaces which millions of people worldwide visit and make purchase every second, the shopping experience and competition between these platforms have been significantly changed and recommendation systems have become a more critical part of these platforms and gained popularity in the literature. One of these online marketplaces, in which the recommendation system plays a key role, is second hand platforms. In addition to general recommendation problems, these platforms have several problems which are specific to this domain such as compromising extremely unique item sets that makes the problem difficult with respect to other domains. In this study, we propose two staged model pipelines using state-of-the-art NLP techniques word2vec and paragraph2vec to address these problems with high quality personalized product recommendation in a scalable architecture. The model performance is evaluated on both offline experiments which are conducted on historical user clickstream dataset that is gathered from a popular second hand platform and A/B test on a production system. As a consequence of these experiments, the proposed model outperforms the baseline collaborative filtering-based models with respect to selected metrics, in addition, provides significant uplifts on several business metrics in the product system.
SIMDify :|framework for application specific SIMD-processing with RISC-V scalar instruction set
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Şarkışla, Mehmet Alp.; Yurdakul, Arda.
Most of the hardware accelerators communicate with the processor via custom instructions. Since custom instructions are not standardized, each accelerator requires a di erent compiler and user code, which can be a tedious process for the user. To reduce the user burden, we propose a parallel programming framework called SIMDify, which generates single-instruction-multiple-data (SIMD) processors that can achieve SIMD processing without using custom instructions. SIMDify takes an application machine code compiled for scalar RISC-V ISA and simulates it to determine the SIMD processing regions. Then, SIMDify con gures and generates the application-speciffic SIMD processor that executes scalar RISC-V instructions concurrently on the SIMD datapath. SIMD processor consists of a single master and multiple slave processing elements (PE). Slaves focus on SIMD level tasks, whereas the master is responsible for the central control. Proposed architecture is the first SIMD capable RISC-V processor designed in HLS and can operate with a faster clock frequency than the existing SISD RISC-V HLS cores. SIMDify relieves the user from using custom instructions with rigid programming models and o ers a exible solution. The processor is designed and tested in Vivado High Level Synthesis 19.2. It operates at 78 MHz on Zynq Zedboard FPGA. Master PE uses 5% and each slave uses 3.5% of FPGA resources. Test results show that execution time can be improved by 8.5x with 9 slaves and 19x with 29 slaves.
Control of inverted pendulum:| comparison of various strategies for swing up and balancing
(Thesis (M.S.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2005., 2005.) Er, Fatih.; Istefanopulos, Yorgo.
In this thesis, two different control schemes are designed for swinging up and three different schemes for balancing of an inverted pendulum, a non-linear electro-mechanical system with open loop unstablity. The controllers which are implemented for experimental purpose are a heuristic control technique using fuzzy logic and an energy control technique for swinging up. In order to balance the pendulum in the upright equilibrium, state feedback control, LQR control which is an optimal control technique and well known traditional control technique PID control are implemented. The control objective is swinging up the pendulum from its initial downward equilibrium to the upward and then balance at this state while it is tracking of desired trajectory. Three different combinations of proposed control schemes are chosen and their performances are compared with each other. Having compared all combinations, it is observed that every controller has advantages and merits for different cases.
Improved handling of SMS messages with statistical natural language processing techniques
(Thesis (M.S.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2005., 2005.) Yıldırım, Ömer.; Say, Ahmet Celal Cem.; Arslan, Levent M.
The Short Messaging Service (SMS) is built on the ability of mobile telephones tosend and receive text messages. SMS based applications are increasing dramatically day byday in the telecommunications industry. The most common use of SMS is for notifying mobile phone users that they have new voice or fax mail messages waiting. Whenever anew message is dispatched into the mailbox, an alert by SMS informs the user of this fact.The Short Message Service can also be used to deliver a wide range of information tomobile phone users from share prices, match scores, weather, flight information, news headlines, lottery results, jokes. In general, user interaction based SMS services requestsome predefined keywords from the users and respond to them after processing theirmessages.However, most users think that they are communicating not with a machine but withhumans, so they compose misspelled and/or machine specific messages containing more than just the needed keywords. As a result, they receive error messages from the server andgenerally do not continue to use the software after trying two or three times by makingsame mistakes. In this thesis, I introduce a new Short Message Service (SMS) parsing model usingStatistical NLP Techniques, whose aim is to solve the existing SMS user subscriptionproblem of a real software company. To do this, the N-Gram statistical approach will beused.
Automated requirements classification using feature selection based on linguistic features
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Çevikol, Sercan.; Aydemir, Fatma Başak.
Requirements classification is an important problem in organizing the systems and requirements, and it is widely used in handling large requirements data sets. A basic example of a requirements classification problem is the distinction between the functional and non-functional (quality) requirements. The state-of-the-art classifiers are most effective when they use a large set of word features such as text n-grams or part of speech n-grams. However, as the number of features increases, it becomes more difficult to interpret the approach, because many redundant features have to be explored that do not capture the meaning of the requirements. In this study, we propose the use of more general linguistic features, such as dependency types, for the construction of interpretable machine learning classifiers for requirements engineering. Through a feature engineering effort, assisted by tools that interpret graphically how classifiers work, we derive a set of linguistic features. While classifiers that use the proposed features fit the training set slightly worse than those that use high-dimensional feature sets, this approach performs generally better on validation data sets and is more interpretable. We use industry data sets, and we perform experimental runs using several automated feature selection algorithms to explore whether our feature set can be optimized further using one of the automated selection algorithms. Although in some data sets, impressive results were obtained. the automated selection algorithms did not prove a significant improvement, and even, on average, the results were worse than the results we obtained using the set based on linguistic features.
Targeting in chaos using analytically described clusters
(Thesis (M.S.) - Bogazici University. Institue for Graduate Studies in Science and Engineering, 2002., 2002.) Sütçü, Yağız.; Denizhan, Yağmur.
The OGY method provides a simple but powerfbl approach of controlling chaotic dynamics. This method can stabilise inherently unstable equilibrium modes of dissipative chaotic systems under the lack of knowledge about the system equations. However, it has the typical drawback of a long waiting time until the system starting from random initial conditions enters the close neighbourhood of the equilibrium mode to be stabilised, where the controller can be activated. The reduction of this drawback is known under the name of targeting. The Extended Control Regions method is a targeting approach, which can operate under the Iakof knowledge about the system equations by employing local models of the system dynamics extracted from empirical data. The method is based on the idea of identifiing and modelling those regions of the phase space, starting fi-om which the system can be steered to a close neighbourhood of the target within a few steps applying sinall perturbatiotns in the control parameters. So far, the modelling of the system dynamics within these phase space regions have been realised using artificial neural networks. In this study, two different strategies are developed in order to realise the clustered version of the Extended Control Regions method on basis of simple analytical models rather than neural networks. Each cluster obtained the gathered data is analytically described as a hyper-ellipsoid. Subsequently, the analytical models of the clusters are used for targeting purposes by applying small discrete variations in the control parameter. Simulation results on several chaotic systems with single control parameter show that the proposed method can achieve targeting using less memory and computation time than the Clustered Extended Control Regions method on cost of a slower targeting performance.
Depth-based scene mapping through spatio-temporal knowledge integration
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Durukan, Meriç.; Bozma, H. Işıl.
This thesis is concerned with scene mapping by a mobile robot using point cloud data. It is a complex process that requires the robot to segment the incoming data, represent it compactly and e ciently, and then use the resulting knowledge in its learning and decision-making. Segmentation enables the robot to determine the point cloud object candidates. The robot bases its learning and reasoning on the detected segments. Range sensors, such as LIDAR, are essential for a robot to extract environmental information. However, they generally create sparse data. For this reason, the sparse data should be considered specially. A novel approach to segmentation is proposed based on an extension of density-based clustering in the spherical coordinate system. We present the deformable sphere approximation (DSA) descriptor as a novel 3D descriptor that encodes point cloud objects. Experimental results show that our representation method is capable of classifying the objects. Finally, we consider how the robot can use all knowledge available to it. We propose an approach in which the robot also considers the knowledge accumulated through tracking the objects' temporal continuity. For this, we propose the temporal deformable sphere approximation (TDSA) descriptor. Its construction requires the robot to track object candidates. For this, we propose a novel multi-tracking approach based on combining Kalman Filtering and multi-object matching considering position and shape similarity. We then compare the various schemes the robot can use in order to utilize the resulting knowledge. Our experimental results show that the T-DSA descriptor improves the classi cation performance compared to only the instantaneous DSA descriptors. As such, the robot is able to build and evolve a scene map as it is navigating in it.
Accessibility on the Web through semantic and social renarrations
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Hoş, Emre.; Denizhan, Yağmur.; Üsküdarlı, Suzan.
The Web was proposed in 1989 with the vision of being an open platform that everyone can access. This vision of providing accessibility to all people is one of the primary goals of the Web. To this end, the (WAI) was launched by W3C in 1997 mainly to improve the accessibility of Web for the people facing physical accessibility barriers. While such work is crucial, it falls short in delivering true accessibility, since further barriers to understanding the information exist. Such barriers are associated with the ability of a person to understand the information that reaches them, such as language, literacy, culture, and expertise. To overcome such barriers, this work utilizes the renarration method, the process of forming an alternative version for a pre-existing web document to reach alternative audience. Accordingly, this work proposes an ontology that represents renarrations, web documents, and social interactions of renarrators and renarrations, and interconnections between them to support creation of an ontology driven social renarration platform. A prototype that utilizes Solid, a web decentralization project, has been created to demonstrate the feasibility of the social renarration approach and the suitability of the ontology for supporting the realization of a decentralized social renarration platform that protects the provenance and privacy.
Classification and detection of wheezes in respiratory sounds
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Şerbetçi, Çağlayan.; Kahya, Yasemin.; Şen, İpek.
Analyzing respiratory sounds and detecting anomalies in them with intelligent computer algorithms has opened a new era for auscultation that has 250 years of history. These algorithms can overcome the drawbacks of conventional stethoscopes and support medics about auscultation. In this thesis, a new intelligent algorithm to detect wheezes superimposed on vesicular sounds is developed and presented. Detection of wheezes with intelligent algorithms is one of the hot topics currently being researched by many researchers. They are continuous musical adventitious respiratory sounds. Their duration, intensity, and phase in respiratory sounds give essential information for the diagnosis and prognosis of respiratory diseases. In this study, one of the aims is to determine the best discriminative features among nine features which are mostly used in other researches. The other aim is to find the best-performed machine learning classifier to classify wheezes and normal respiratory sounds. Last, we created a novel detection algorithm is presented to detect correctly the wheeze interval in recorded respiratory sounds by employing selected machine learning model to respiratory sounds.
Authorship recognition in online social platforms
(Thesis (M.A.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2017., 2017.) Kuzu, Rıdvan Salih.; Salah, Albert Ali.
Biometrics is the identi cation of a person by personal properties and traits, and can be divided into physiological based and behavioural based methods. In this thesis we investigate the identi cation of users of a social platform from their verbal behaviour, which is an example of behaviour based biometrics. Online social platforms implement moderation mechanisms to lter out unwanted content and to take action against possible cases of verbal aggression and abuse, sexual harassment, and such. Since they can have large numbers of users, it is desirable to automatize parts of this process. What we call chat biometrics aims to re-identify a user from chat messages. The typical application scenario is the re-identi cation of banned users, returning under di erent identities, and aggressors operating through multiple fake accounts. We propose a processing pipeline, and contrast the problem with the authorship identi- cation problem, which is well-studied in the literature. We evaluate our proposed approach on a large corpus of multiparty chat records in Turkish (namely, the COPA database), which was collected from a multiplayer game environment. We also introduce a new corpus in this study, collected from a well-known Turkish social platform called Ek sisozluk, in order to test the robustness of the system across domain changes, as well as on Portuguese and English news datasets, to show performance across languages. We evaluate both pro le-based and instance-based approaches, and provide detailed analyses with regards to the required amount of text to identify a person reliably.
Human-like coordination of body-assisted arm movements for object manipulation
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Cumali, Kadir.; Bozma, H. Işıl.
Manipulation is an integral capability for service robots. The goal of this thesis is to design and develop an approach that enables a mobile robot to mimic human manipulation abilities. We consider a diﬀerential type of mobile robot that is endowed with an arm and gripper. The robot is assumed to have visual sensing so that it can determine the relative position of the object of interest. First, it is observed that hu mans exhibit various basic modes of interaction with an object of interest, including extension, ﬂexion, gripping, release and translation. As such, the robot can be pro grammed to have similar capabilities through establishing the correspondence between the robot and a human with respect to the underlying manipulation mechanisms. More complex behaviors such as putting, pulling, pushing, and shaking are deﬁned using a sequential composition of basic operations. Second, humans are observed to achieve these tasks through the coordination of their body and arm movements. For this, a control approach in which the movements of the robot body and manipulator are cou pled temporally and spatially is proposed. As such, if the object of interest is within the robot’s reach, then only arm movements are made. If this is not the case, the robot starts moving its body. Depending on the vicinity of the object, this may be accom panied by arm motion or not. The control algorithm results in the robot’s body and arm movements to be done in a coupled manner. The proposed approach is evaluated through an extensive set of experiments involving various manipulation tasks.
Deep learning based text regression
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Dereli, Neşat.; Saraçlar, Murat.
Most ﬁnancial analysis methods and portfolio management techniques are based on risk classiﬁcation and risk prediction. Stock return volatility is a solid indicator of the ﬁnancial risk of a company. Therefore, forecasting stock return volatility success fully creates an invaluable advantage in ﬁnancial analysis and portfolio management. While most of the studies are focusing on historical data and ﬁnancial statements when predicting ﬁnancial volatility of a company, some studies introduce new ﬁelds of information by analyzing soft information which is embedded in textual sources. Fore casting ﬁnancial volatility of a publicly-traded company from its annual reports has been previously deﬁned as a text regression problem. Recent studies use a manually labeled lexicon to ﬁlter the annual reports by keeping sentiment words only. In or der to remove the lexicon dependency without decreasing the performance, we replace bag-of-words model word features by word embedding vectors. Using word vectors increases the number of parameters. Considering the increase in number of parame ters and excessive lengths of annual reports, a convolutional neural network model is proposed and transfer learning is applied. Experimental results show that the convolu tional neural network model provides more accurate volatility predictions than lexicon based models.
State - space modeling of a planar solid oxide fuel cell
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2016., 2016.) Demirezen, Gülsün.; Denizhan, Yağmur.; Demircan, Oktay.
In this thesis, state-space modeling approach is applied to a complete planar Solid Oxide Fuel Cell (SOFC) system. Several simulations of the dynamic system have been performed for 400°C to 800°C as operating temperature. In addition, the results of the simulations are compared with experimental results obtained at Fuel Cells and Energy Laboratory at Boğaziçi University. The validity of the model is checked through these comparisons. In this study, the performance of a planar SOFC system is investigated on basis of simulations obtained on a MATLAB® software platform by using dynamic modeling approach. This study allows us to observe influence of various material parameters on the operation of the planar SOFC and to determine the limiting factors of mechanism considered. In addition, the model supports the interpretation of the experimental results obtained at Fuel Cells and Energy Laboratory at Boğaziçi University.
Application of variable structure systems theory for training of intelligent systems
(Thesis (M.S.) - Bogazici University. Institue for Graduate Studies in Science and Engineering, 2001., 2001.) Yıldıran, Uğur.; Kaynak, Okyay,
Soft computing architectures with their extensive flexibility and strong mapping capabilities have been widely used for control of nonlinear systems. In this regard, error backpropogation and its derivatives have been the most popular and frequently employed schemes for parameter adjustment of these architectures. However, these schemes bring some serious problems together, like instability of closed loop system and sensitivity to uncertainties, which must be carefully addressed by a system designer. In order to alleviate these problems, recently, Efe has proposed a control strategy in which parameters of intelligent controllers are updated by a continuous-time robust parameters adjustment mechanism in order to robustify and stabilize the closed loop system dynamics. The results obtained for a two link SCARA robot in this study show that the proposed method is successful in achieving the control objectives. In this thesis, the methodology proposed by Efe is investigated for first order nonlinear systems. Based on the results, it has been observed that the time evolution of input-output curves of different structures show similar characteristics. Moreover, a modification is proposed for update mechanism of all architectures in order to prevent unbounded parameter evolution problem which occurs in the original algorithm. Lastly, based on the results for different systems, it has been concluded that the Adaptive Linear Element is the most suitable architecture for the control systems investigated because of its simplicity.
Seed-based and data-driven analyses of default mode network connectivity measures in dementia
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2017., 2017.) Kılıç, Başak.; Acar, Burak.
Functional neuroimaging and its applications to neurodegenerative diseases and mental illnesses have created an enlarging area of interest that varying lines of research ranging from molecular biology to engineering contribute to. Among them, Alzheimer's disease has a critical importance by causing the largest number of dementia cases. Recently, mild and subjective cognitive impairments have also been associated with Alzheimer's as possible indicators of cognitive decline. Using resting-state fMRI to investigate functional connectivity measures and detect any abnormality within and between networks have yielded promising results that disclose information about the nature of the diseases. The objective of this thesis is to use varying resting-state fMRI methods to di erentiate between SCI, MCI and AD patients by investigating the changes within Default Mode Network (DMN). The obtained results indicate that the changes within the functional connectivity measures among DMN components can be detected independent of the method of choice, and the measures of connectivity di er among groups. Subsequent research would aim for detection of possible bio-markers that are present through several stages and nding a common framework where metrics obtained from di erent methods can be compared.
Investigation of self - organisation in the Benard experiment based on micro-scale simulation
(Thesis (M.A.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2017., 2017.) Kaya, Emre.; Denizhan, Yağmur.
Self-organisation is the phenomenon where in a dynamic system made of autonomous, yet interacting components a global macro-scale regularity observable by an outside observer spontaneously emerges. Rayleigh-B enard Convection is one of the most common examples of this phenomenon. In this thesis, the B enard experiment, which involves the self-organisation of convection cells, has been simulated at micro-scale, i.e. at molecular level, in order to investigate the dynamics underlying this self-organisation phenomenon at its most fundamental level. Molecular dynamics simulations of the proposed 2D micro-scale model have been conducted under di erent external conditions to observe the dynamic behaviour range of the system. An image processing algorithm based on curl of the velocity eld has been developed to automatically detect the presence or absence of convection cells and thus the type of the dynamic regime at hand. The 2D micro-scale model developed in this thesis sheds light on how the dynamic regime depends on external conditions and provides an answer to the original question of this study whether the emergence of macro-scale order can be detected from the micro-scale perspective of a single particle.
3D cow identification in cattle farms
(Thesis (M.A.) - Bogazici University. Institute for Graduate Studies in the Social Sciences, 2013., 2015.) Arslan, Ahmet Cumhur.; Akar, Mehmet.
Animal farms have been steadily growing to meet the consumption requirements of the society in an e cient manner. This fact necessitates new monitoring and tracking systems to collect useful information about the herds in order to observe their general health and instantaneous state. However, recognizing and tracking an animal in a farm is a di cult task due to the target's similarity and hard to predict dynamics. In this thesis, a novel cow identi cation system is proposed. There are prominent features of this solution which di erentiates it from the others in the literature, i.e., it does not need any markers or external devices placed on the animal; works in even unlighted environments; identi es even black cows without distinctive coat patterns; is relatively cheaper, and enables accurate positioning. Proposed solution is based on 3D shape analysis of the top back part of the animals captured with RGBD cameras placed at an adequate height, where two dimensional images are constructed with respect to the local surface features and are subsequently identi ed by using face recognition methods. To evaluate the applicability of the proposed system, a real-time prototype software has been developed and a 3D cattle dataset is acquired which, to our knowledge, is unique in the literature. This dataset is gathered from moving animals which do not have distinctive coat patterns and captured in di erent lighting conditions. Applicability of the proposed solution has been veri ed by testing with the acquired dataset. Convincing results are obtained where %88 of 50 cows are identi ed successfully.

Browse

Recent Submissions