Question:
Discuss about the Applying Kinect to Human Body Height Measurement to Improve Character Recognition Performance.
Answer:
Introduction
The human body tracking comes from the 2D effects where there are challenges related to the problems when related to the vision of the computer. For this, there are different factors for the N rigid body parts, where the entire body is also posing to the representation of the long vectors where there is a representation of the translational and the rotational motion of the different parts of the body. Here, the forms are set with the body pose which are in the higher dimensional continuous space that is not common. The estimates are based on the forms which includes the structure and the higher dimensionality of the state space that will lead to the problems of the intractable computation complexity. (Lee et al., 2015). The effectiveness of the high dimensional forms is for the body pose space where the people tend to use the sampling based methods or the learning methods to work on the different forms of the problems. This is also for the annealed sampling as well as the partitioned sampling that is based on working with the posterior that is related to the forms with the higher continuous space. At the time of tracking, the particles are propagating with the use of the dynamic model which is weighted by the image. There are different basics, important for the sampling and work on the sequential importance where the dimensions of the pose structures and space is inefficient. Inorder to reduce the sampling to the partition, the pose spacing is set into the sub space which is then generating the samples to work on the different number of the required structures. For this, there are forms which works on the reduction parts to handle the computation load to the generation step of the sample. This is then considered to be computationally inefficient. (Sobel et al., 2014).
Background
It has been seen that the learning based methods are set with the attempt to focus on direct mapping from the image feature space to handle the pose space where the mapping has been from the learning of the data and training. The most important learning based methods are set to track the changes and the techniques like the regression methods and the Gaussian process latent variable model.it works on the learning based methods which could be for the specific motions as well as for the training of the data. The forms are robust and the settled with the effective head and the limb detectors to work on the whole-body posture. The body tracking techniques are based on tracking the parts of the body and work on the reduced forms of the problems with higher dimensions. It is also seen that there is a significant standard for the changes in the parts which tend to occur. (Monkarsi et al, 2014).
The focus is also on the adjacent human body parts which are found to relate to the different joints that is set with the muscles. There are other forms of the feasible poses which are for the upper body and that tends to handle the anatomical constraints. The research is about the positions with the constrains to track the problem, with the capturing of the relationships in a proper and systematic manner. A proper probability standard for the graphical model is set with the natural way of the relationships to handle the segmentation which is mainly in between the adjacent parts of the body.
Motivation
There have been researches mainly dealing with the constraints of the body tracking problems where the problem is mainly to effectively handle the relationships in a systematic manner. (Nguyen et al., 2016). For this, there is a need of the natural system which includes the use of thermoballistic graphical methods for the tree-structured growth.
Objectives
The major objectives are set for the inference of the body poses which is for the symmetry of the human body. The work is shown using OpenCV, C++.
Outline
There will be a proper description of how the data is used to track the human body measurements using OpenCV. We are also going to work on the upper and the lower part of the body.
Literature Review
The action related constraints works on the kinematics constraint which is used for the constraint along with the body parts. The two adjacent body parts which is connected through the joints. The connectivity is elastic with the joint points with the two adjacent parts which is smaller to handle the higher probability. (Roncone et al., 2016). The physical constraint includes the imposing with the physical infeasibility pose, along with the non-adjacency parts. The example is set with the 3D body parts that cannot intersect with each other. The biomechanics constraint works on the relationships with the body parts and the kinematic constraints which restricts the motion. The anatomical constraint is also able to provide the statistical data with the distribution of the different body parts dimensions. The appearance constraints are set to pose the different body parts, where the appearances are set to handle the accuracy forms. It includes the detection of the occurrence with the self-occlusion and the automatic reduction of the confidence in the measurement of the different parts. The paper is based on the framework which is for the 3D upper body based on the Dynamic Bayesian Network, where the model tends to capture the spatial and the temporal forms of the relationship in the different body parts. With this, one can also encode the probability of the constraints that include the Conditional Probability Distribution which is for the explicitly models to work on the parametric forms. (Lee et al., 2014). The measurement of the detector is based on the requirements of the data to be generalized in the different motions. For the proper tracking, one can also use the DBN for the incorporation of the temporal relationships.
The work is related to the human body that tend to pose the estimations which are based on the tracking of the different graphical models. This is set through the Factor graph and the Conditional Random Field structure that works on the bottom up approach. It also includes the forms with the independent search that is to find the pose and the parts that could be combined for the generic constraints as well.(Phan et al, 2015). The approaches are based on the tree-structured measures with the undetected forms of the graphical model. It is mainly when the simple tree structures allow the effective inference set with the other forms of the associated models. This is connected to the encoding of compatibility between the different pairs of configuration. Here, the parts are set for the training process to handle the parameter based on the relationships that is set in and among the body parts. It could easily be defined through the training of the data as well as learning model that will not be able to generalize the well unexpected poses. There are different forms of the modeling of the relationships which is connected to the body parts. (Lin et al., 2016). The directed forms of the graphical models are mainly to capture and handle the dynamics which is based on the body pose. The structure is also based on the capturing of the relationships that is mainly for the simple forms of the kinematic constraints of the joints as well as the coordination that sets the relationship in between the limbs. Here, the limbs can coordinate with the action related factors as well as the other forms for the running or walking. The incorporation the 3D joint limits, with the non-self-intersection of the constraints is set with model priors that can define the limits of the angle along with the image matching costs. Here, the structure is based on working towards the constrained optimization using the covariance which is scaled with the sampling followed through the local refined parts as well. It is important to focus on the ambiguity factors which are for the human low body structures inorder to track the changes with the physics based model. The generalization is set with the approach that leads to the motions, and so the self-occlusion is the major problem for tracking of the body. (Madadi et al., 2015). Hence, one can easily notice that there is a need to build the deterministic graph that is for the occlusion visibility factor.
Length Constraints
When there is a need to consider the right arm and right forearm, there are connections which are set with the common joints, where the positioning is based on the right forearm reference. This is close to the estimated position which could be set with the Gaussian distribution process. There are length constraints which are set with the human body that is symmetric. Here, there are systems which works on the length for the left side and counterpart on the right side. The symmetric relationships are set with the modelling structure into the upper body model. There are constraints which tends to hold the 3D model of the pose tracking when there are images which have been for the nearly frontal viewpoint. The enforcement is mainly for the introduction of the additional constraints which is for the elliptical nodes that have been set with the symmetric forms of the relationship. There are nodes for the constrain which is for the Gaussian distribution which is the difference for the parent states. The covariance matrix has been set with the diagonal matrix to handle the entries designed with the constraints on the part lengths where there are no constraints on the part positions.
The adjustments are for the mean and variance which impose the length constraint at the different level which is set for the equal length constraint.(Pfister et al., 2014) The states are set for the large numbers where there is variance of the length set to handle the small constant. There are forms which include the means of the distribution which includes the difference of the parent states. The covariance states are the diagonal matrix with the diagonal set for the difference in between the parent states. The designing is based on the part lengths with no constraints on the part positions. The diagonal matrix is set with the adjustment of the mean and the variance that impose the length of the constraint to the different levels. The process includes the positions that have been set to check on the lengths along with the enforcement of the symmetry constraints on the length. The pattern is based on working over the evidence coding where there are states set for the zero mode to enforce the symmetry constraint. The forms of the variance are set with the length l which is small. For this, the lengths of the parent nodes are for the encoding of the symmetry functions and the constraints on the length. (Blomgren et al., 2015). There is no major relationship in the x and the y direction as their variances are large. Here, the symmetry constraints are set based on the length and the forms where the constraint nodes are for the evidence node patterns. The forms of zero are set to handle the symmetry patterns along with the variance in the length that works on setting the demands with the encoding depending upon the symmetry patterns which includes the anatomical constraints. The parts of the human body are proportional to the length with the relative forms set with the ratio of the adjacent forms. Here, the parts are set with the constant value, to check on the relationship which can easily be set to make sure that there is a proper tracking of the upper body parts with the computability of the lengths. The forms are set with the relationship that includes the body structure parts with the forearm encoding that allows the physical relationship with the anatomical relationship. It is for the right arm system that works on the relative standards with the right forearm length to the right arm length which could easily be understood from the anthropometric data.(Phan et al., 2015).
Occlusion Modelling
This includes the use of the problems in the upper body tracking where there is an uncertainty with the image measurements that include the occluded parts that tends to increase with the major contribution to the tracking which could easily be reduced. This is based on the forms with the upper body model that has been set with the occlusion of the head by the forearm. Here, the standards are based on the modelling into the BN model which is mainly to track the parts with the left right forearm in detail. The upper body model is set with the occlusion model where there are additional forms of the binary models that have been set with the BN to handle the setup of the right forearm that includes the measurement of M1 and M2. The measurements are based on the independent tracking of the head with the right forearm and the left forearm setup. Here, the distance is set in between the detected head and the right forearm which includes the half size of the head as well. The links are directed in between the nodes and the measurements are also based on encoding the uncertainties for the measurements. The conditional probability is based on representing the empirical sets which influence the head and the occluded forms of the either parts. Here, the image is set for the results that works on the lesser accurate functions. (Loper et al, 2014). The modeling is also based on working with the variances as well as the forms that tend to include the head measurement. The increase in the variance structures of the Gaussian system is set when the occlusion tends to happen. For this, there is also a similarity to the influence based on the forearms structures. The modeling is based on working over the uncertainties that is set with the variances of the CPD structures.
Extension to 3D Pose Tracking
It includes the approaches of the 3D setup where the upper body is mainly to handle the tracking from the stereo sequences. Along with this, there is a state of the body which includes the six parameters. Here, the forms are set to check on the 3D positioning of the reference points, with l being the length of the body part the structure of the model tend to remain same with the need to change on the CPD mainly due to the state parameters. There is a clustering that is set with no major strong relationships in between the angles of the adjacent body parts. This is mainly due to the freedom of the motion.
Tracking with Bayesian Network
The extensions are for the dynamic Bayesian network where one can easily track the forms with consideration of the temporal smoothness of the motion. The assumption is based on the modeling where the structures are set to check on the stationary forms of the Markov chains. The tracking of the problems as well as analysis of the inference problems in the Bayesian network is to check on representation of the temporal evolution. In the model, there are parts which are for the each of the upper body part structures to handle the previous frames and the parent nodes that have been set at the BN upper body model. (Jampani et al, 2015). The structures are also suitable for the modelling based on the linear Gaussian system setup. Here, there are forms with the two parts which include:
- The RF which is for the representation of the mean values as well as the temporal dynamics.
- There is other part which is set with the representation of the estimated mean value with the relationship that is set in between the body parts. The constant is to balance the influence with the focus on the temporal dynamics and the constraints mainly in between the parts of the body.
- Here, the selection is also based on the changes with the form of tracking that denotes all the important measurements in the frame that is set to model the inference problem.
The structure has been set with the modelling with the upper body tracking of the system. It mainly includes the temporal dynamics that has been set in between the hidden forms of the state variables. There are dotted links which are in between the pair of the state variables that is set for the different lines of illustration. The focus is on the measurement forms with the model that includes the frames with the DBN model.
Measurement Extraction
The extraction is based on the part measurement with the performance that includes the DBN performance based on the prediction with the estimation that includes the process to measure the forms with the automatic forms which includes the 3D tracking. (Chong et al., 2016). The hypothesis is based on the forms where there are 3D parts that are projected to handle the projects with the right and the left images extraction of the right forearm regions. The sum of the squared distance is based on the criteria with the right and left images that includes the tracking of the results using the occlusion nodes. It works on the forms with the tracking of the system with the occlusion nodes. This is set mainly for the effects which includes the upper body model with the length constraints to keep a certain relative relationship for the body parts. Here, there are different effects which are set for the DBN tracking model, without any use of the length constraints.
Implementation and Testing
The work is done in Open CV which is the best library for handling the computer visions of the systems. This is based on the cross platform which is free to use under the BSD standards. Here, there are forms to handle the real-time ray tracking and the 3D display of the walls where the contributors need to check on the optimization of the experts along with working over the project goals. (Kaehler et al., 2014). With this, there is also certain advancement based on the system provisions that is not open but is set with the higher codes where there is no reinvention of the wheel. The dissemination of the vision knowledge is through the common infrastructure with the developers who can easily build up the readable formats. Here, the advancement is set for the vision based applications by making it completely portable with the performance based structured that are for the setup of the codes in effective manner. The software is based on the types of the patterns and the structures where the implementation is for the performance on the multi-core systems. The performance is also including the 2D and the 3D kits where the estimations are depending upon the understanding of the motion, identification of the object and the segmentation process in effective manner. The forms are set with the boosting and the decision tree learning that is able to handle the gradient boosting of the trees with the maximization of the expectation of the algorithm. (Shrivastava, 2013). The code’s idea is to convert the images into the person silhouette and using pixel intensity values, measure the width of a person’s neck, arms, waist, thighs.
1st: obtain silhouette of person/ detect edge of person
2nd: with a black and white silhouette for example attached below.
3rd: using multiple loops, we go through the pixels from left to right of the image and detect the different body parts and measure the body parts.
There is a possibility of the system to work on the processing of the image mainly through the vision applications that has been set with the computer vision applications. The image processing is set with the relevant forms of the information from the images to make the decisions that will help in obtaining all the important information. (Laganiere, 2014) There are other image manipulation and the decision-making factors which are based on the computer vision applications that tend to handle the hardest ways for the setting of the methods of Open CV and C++. The standards are set with no major documentation or the error handling of the codes. This works on open source of the C++ library with the processing of image in the form that includes the development of the commercial and the non-commercial patterns. There are forms where the applications are aiming for the different times of the development with the computer visions applications that is easy and efficient. The key features are for the optimization of the real-time applications with the real image processing and checking on the computer vision applications. The standards are set with the full interface planning, where the Open CV could easily run into the applications on Windows, Android and the Linux. There are modules set for the structural development, which provides the links that point to the core of the data structures, with the basis of the image processing functions. (Guttler et al., 2016). Here, there are modules like the high-up. The forms are set with the algorithm that include the filtering of the image, transformations and the color space conversions. This is for the object tracking algorithms, background subtraction of the algorithm etc. the detection and the recognition of the algorithm is for the different standard objects.
The image filtering is important for the computer vision, where the applications are set to handle the filtering of the methods into the images. The forms are set with the eroding, dilating, inverting the methods which are mainly to detect the shapes. The shape detection and the tracking is mainly through the contours which could easily be used for the setup of the different points on the vertices. This includes the identification of the polygon with the different number of the vertices that have been set to identify the features of the polygons like the complexity and the concavity of the system. Here, the major effect has been on the length constraints that is set with the forms of the different parts. Here, the modelling procedures are set with the proposed model which could easily set the track for the torso structure model with the relative accuracy for the other parts. There are tracking parts set for the arms and the forearms where there is a proper moving of the fast and the wider versions to track. There is no major drift in any of the sequences. (Aoki et al, 2016). Here, the 3D upper body tracking is mainly depending upon the tracking that could be tested with the two stereo type sequences with the frames and the setup of the body motions. There is a proper manual labelling that is through the obtaining of the ground truth with the sustainability of the quantitative performance or the evaluation. The work is in OpenCV which includes the use of the 3D pose tracking results with the quantitative evaluation to handle the setup with the tracking of ends in effective manner. The quantitative evaluation is also based on the forms with the 3D upper body with which there is also the addition of the model with the occlusion structures. There are demonstrations based on the modelling standards with the approach to combine with the C++ language. ( Takahasi etal., 2013).
The work is in C++ which is for the generalized programming language that is able to set the different levels of the lower level memory manipulation. This has been set with the system performance along with the highlights that works on the software infrastructure and the resource applications which includes the desktop, server and the performance critical applications. The setup of the features is set with the actual problems and the features that is able to relate with the real-world problems along with working on the implementation of the useful features that is important to prevent the system from the misuse of C++. (Tsahev, 2013). There is a proper allowing of the feature that will lead to the facilitation of the system with the well-defined network facilities that is for the combination processes with no major parts development. The user created types are then to work on the actual problems and the features which need to work on the implementation of the prevention of every possible misuse of the C++. This also leads to the facilitation of the organizing programs as well as the well-defined standards to provide a better support to the system performance with the in-built types. The structures are also related to the forms which includes the negative impact that is created in the lower performance, where there are functions working with the static storage duration of the objects and the initialization of the objects in the two phases. This is formed through the factors with the zeros that related with the standards setup with the comping of the time and then saving the data partition in the executable format. This is also done through the constructor or the function call. (Archieves, 2015). The automatic storages are mainly for the system to work on the factors which handles the system processing with the points to declare the local variables which are then set with the local block variables. The forms are set with the member variables when the parent object is created. This is set with the array members who tend to initialize from 0 to the last member of the array in the order. For this, there is a reverse order of the creation that has been set with the temporary variables which are created with the expression of the evaluation and the destruction when the statement contains the evaluation at the end of the statement. The forms are set with the assurance of the data structures and the operators which are for the intended setup the usage of the model for the developer. (Lyons et al., 2015). The class of the members are set with the primary encapsulation process with the mechanisms that works on the public, protected and the private data connection to enforce the encapsulation. The standards are for the system to handle the functions access along with the other forms which could easily be able to support the C++ along with the entities that are not the part type of the representation. It works on the modular programming which is set for the functions that are mainly the part of the minimal system interface for the user of the class. This includes the implementation of the details as well as allowing the user to work on the fundamental development without any major change in the system designing. The work is based on the acquiring of the properties of the data which includes the declaration of the public, protected or the private setup which works on the forms that are to ensure that there is a one instance for the base class that exists in the inheritance graph, thereby, avoiding the ambiguity of the multiple inheritance problems.
Testing
The unit testing is based on the components where the software could easily be tested. For this, one need to work on the validation that the units are set for the programs to design. The procedural programming includes the individual program, function and the procedure that has been set with the practice response to the smallest unit methods which belongs generally to the base class or the super class. (Nicolai et al, 2015). The unit testing is for the drivers, stubs and the objects. The plan is mainly based on the performance of the software developers and to work on the unit testing plan for the preparation, reviewing, working and handling the baseline concepts. The confidence to change and work on the time when the code is changes check on the formats with the unintended impacts to the code. The codes are also reusable with the modularity. The development is fast which works for the performance and the fuzzy development test to check on the writing of the tests and taking the time which is compensated by the lesser amount of the time which takes it to run the tests. It works on the development with the cost of fixing the defect detected at the higher levels. The forms of the codes are reliable where debugging is easy when the test fails or changes.
Integration Testing
The setup is based on working over the performance with the exposure to the defects in the interface. For this, the interactions are based on the components and the systems, which works on the exposure patterns with the system integration testing. The forms work on the interface to the external organization, to handle the analogy of the tasks that depend on the preparation, review, and the reworking of the system in effective manner. The complete assurance is mainly related to the detailed designing of the document where the interactions have been set with the units clearly defined to work towards the integration testing without any information. (Hansen et al., 2015). The lack of the integration testing is for the software modules which are combined and tested to work on the system efficiency. It also works for the different forms of the access to the system that could help in the larger aggregation to the applied tests and the defined integration tests plans. It can deliver the output as per the system testing formats, where there is a collaboration integration, layer integration and the client server integration. The approach is for the bottom up testing which is for the integrated standards that work on the facilitation of the testing with the higher-level components that have been set. The process is repeated till the component can meet the higher level of hierarchy.
Conclusion and Recommendations
The work is about handling the frameworks in the 2D and the 3D structures where there are dynamic patterns which includes the structural modelling procedures with the physical constraints that include the explicit forms of the modelling structures. The representation is mainly based on handling the model with the different forms of generic streams that are for the anatomical constraints. (Fu et al., 2013). The uncertainties tend to change with the image measurements. It works on handling the effective parts with the standards set to handle the different parts of measurements. There are different options to handle the issues related to the tracking of the object. For this, the involvement of the system to handle the tracking of the object with the adaptation in the online format for the better fit of the scenario. Here, the system is based on the resources which is generally for the online target model with the learning and the adaptation. Here, the forms are set to check on the using of the system functions with the pre-trained or the parameterized model to handle the structure of the object. Along with this, the objects are working with the tracking performance, where there is a need of checking of the online learning and the adaptation of the system. This works mainly with the forms, which includes the online learning and the adaptation to the open issues. This works with the body pose tracking systems which works for the forms that is important for the improvement of the body processes. Here, the efficiency is based on working over the advantages with the integration of the same framework.
References
Lee, H.W., Liu, C.H., Chu, K.T., Mai, Y.C., Hsieh, P.C., Hsu, K.C. and Tseng, H.C., 2015. Kinect Who’s Coming—Applying Kinect to Human Body Height Measurement to Improve Character Recognition Performance. Smart Science, 3(2), pp.117-121.
Monkaresi, H., Calvo, R.A. and Yan, H., 2014. A machine learning approach to improve contactless heart rate monitoring using a webcam. IEEE journal of biomedical and health informatics, 18(4), pp.1153-1160.
Sobel, D., Kwiatkowski, J., Ryt, A., Domzal, M., Jedrasiak, K., Janik, L. and Nawrat, A., 2014, September. Range of Motion Measurements Using Motion Capture Data and Augmented Reality Visualisation. In International Conference on Computer Vision and Graphics (pp. 594-601). Springer International Publishing.
Lee, E.C., Whang, M., Ko, D., Park, S. and Hwang, S.T., 2014. A New Social Emotion Estimating Method by Measuring Micro-movement of Human Bust. In Industrial Applications of Affective Engineering (pp. 19-26). Springer International Publishing.
Nguyen, D.T. and Park, K.R., 2016. Enhanced Gender Recognition System Using an Improved Histogram of Oriented Gradient (HOG) Feature from Quality Assessment of Visible Light and Thermal Images of the Human Body. Sensors, 16(7), p.1134.
Roncone, A., Hoffmann, M., Pattacini, U., Fadiga, L. and Metta, G., 2016. Peripersonal Space and Margin of Safety around the Body: Learning Visuo-Tactile Associations in a Humanoid Robot with Artificial Skin. PloS one, 11(10), p.e0163713.
Jampani, V., Nowozin, S., Loper, M. and Gehler, P.V., 2015. The informed sampler: A discriminative approach to bayesian inference in generative computer vision models. Computer Vision and Image Understanding, 136, pp.32-44.
Madadi, M., Escalera, S., Gonzalez, J., Roca, F.X. and Lumbreras, F., 2015. Multi-part body segmentation based on depth maps for soft biometry analysis. Pattern Recognition Letters, 56, pp.14-21.
Phan, A. and Ferrie, F.P., 2015, May. Towards 3D human posture estimation using multiple kinects despite self-contacts. In Machine Vision Applications (MVA), 2015 14th IAPR International Conference on (pp. 567-571). IEEE.
Lin, H.Y., Hsueh, Y.L. and Lie, W.N., 2016, December. Abnormal Event Detection Using Microsoft Kinect in a Smart Home. In Computer Symposium (ICS), 2016 International (pp. 285-289). IEEE.
Pfister, T., Simonyan, K., Charles, J. and Zisserman, A., 2014, November. Deep convolutional neural networks for efficient pose estimation in gesture videos. In Asian Conference on Computer Vision (pp. 538-552). Springer International Publishing.
Nguyen, V.A., Le, T.H. and Nguyen, T.T., 2016, December. Single camera based Fall detection using Motion and Human shape Features. In Proceedings of the Seventh Symposium on Information and Communication Technology (pp. 339-344). ACM.
Loper, M.M. and Black, M.J., 2014, September. OpenDR: An approximate differentiable renderer. In European Conference on Computer Vision (pp. 154-169). Springer International Publishing.
Blomgren, S. and Hertz, M., 2015. Facing the differences between Facebook and OpenCV: A facial detection comparison between Open Library Computer Vision and Facebook.
Chong, P. and Tay, Y.H., 2016, December. A novel pedestrian detection and tracking with boosted HOG classifiers and Kalman filter. In Research and Development (SCOReD), 2016 IEEE Student Conference on (pp. 1-5). IEEE.
Laganière, R., 2014. OpenCV Computer Vision Application Programming Cookbook Second Edition. Packt Publishing Ltd.
Kaehler, A. and Bradski, G., 2014. Learning OpenCV. O’Reilly Media, Inc..
Sobral, A., 2013, June. BGSLibrary: An opencv c++ background subtraction library. In IX Workshop de Visao Computacional (WVC’2013) (Vol. 7).
Kaehler, A. and Bradski, G., 2016. Learning OpenCV 3: Computer Vision in C++ with the OpenCV Library. ” O’Reilly Media, Inc.”.
Shrivastava, R., 2013, February. A hidden Markov model based dynamic hand gesture recognition system using OpenCV. In Advance Computing Conference (IACC), 2013 IEEE 3rd International (pp. 947-950). IEEE.
Shrivastava, R., 2013, February. A hidden Markov model based dynamic hand gesture recognition system using OpenCV. In Advance Computing Conference (IACC), 2013 IEEE 3rd International (pp. 947-950). IEEE.
Güttler, J., Georgoulas, C. and Bock, T., 2016, April. Contactless fever measurement based on thermal imagery analysis. In Sensors Applications Symposium (SAS), 2016 IEEE (pp. 1-6). IEEE.
Gómez, M.J., Castejón, C., García‐Prada, J.C., Carbone, G. and Ceccarelli, M., 2015. Analysis and comparison of motion capture systems for human walking. Experimental Techniques.
Jakubowski, K., Eerola, T., Alborno, P., Volpe, G., Camurri, A. and Clayton, M., 2017. extracting coarse Body Movements from Video in Music Performance: a comparison of automated computer Vision Techniques with Motion capture Data. Frontiers in Digital Humanities, 4, p.9.
Tashev, I., 2013. Kinect development kit: A toolkit for gesture-and speech-based human-machine interaction [best of the web]. IEEE Signal Processing Magazine, 30(5), pp.129-131.
Aoki, K. and Aoyagi, N., 2016. Mark Projection on Real World with Precise Measurement of Angular Velocity for Helping Picking Works. In Transactions on Computational Science XXVIII (pp. 23-44). Springer Berlin Heidelberg.
Fu, Y. and Long, X., 2013. Pedestrian Tracking Based on Improved Particle Filter Under Complex Background. In Advanced Materials Research (Vol. 756, pp. 4103-4109). Trans Tech Publications.
Takahashi, M., Fujii, M., Naemura, M. and Satoh, S.I., 2013. Human gesture recognition system for TV viewing using time-of-flight camera. Multimedia tools and applications, 62(3), pp.761-783.
Archives, A., 2015. 9 Archiving the Databody: Human and Nonhuman Agency in the Documents of Erkki Kurenniemi. Writing and Unwriting (Media) Art History: Erkki Kurenniemi in 2048, p.125.
Lyons, D.M., MacDonall, J.S. and Cunningham, K.M., 2015. A Kinect-based system for automatic recording of some pigeon behaviors. Behavior research methods, 47(4), pp.1044-1054.
Hansen, M.F., Smith, M., Smith, L. and Hales, I., 2015. Non-intrusive automated measurement of dairy cow body condition using 3D video.
Nicolai, P., Raczkowsky, J. and Wörn, H., 2015, July. Continuous pre-calculation of human tracking with time-delayed ground-truth: A hybrid approach to minimizing tracking latency by combination of different 3D cameras. In Informatics in Control, Automation and Robotics (ICINCO), 2015 12th International Conference on (Vol. 2, pp. 121-130). IEEE.
Tseng, S.H., Chao, Y., Lin, C. and Fu, L.C., 2016. Service robots: System design for tracking people through data fusion and initiating interaction with the human group by inferring social situations. Robotics and Autonomous Systems, 83, pp.188-202.
Hsieh, W.F., Chen, L.H., Hung, H.M., Sato-Shimokawara, E., Takama, Y., Yamaguchi, T., Wu, E.H.K. and Chen, Y.W., 2015. An intelligent decision supporting system for international classification of functioning, disability, and health. Visualization in Engineering, 3(1), p.11.
http://www.filehosting.org/file/details/663366/Submit_2.zip
