Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

A Comparative Study of Human Motion Capture and Computational Analysis Tools Seung-kook Jun†, Xiaobo Zhou†, Daniel K. Ramsey‡ and Venkat N. Krovi† † Mechanical & Aerospace. Engineering, SUNY at Buffalo, Buffalo, NY 14260 USA ‡ Exercise & Nutrition Science, SUNY at Buffalo, Buffalo, NY 14260 USA

Abstract Human motion-capture and computational analysis tools have played a significant role in a variety of product-design and ergonomics settings for over a quarter-century. In moving beyond traditional kinematic (and its dual-static) settings, advances in biomechanics and multibody dynamics have led up to computational analysis tools that can provide significant insights into the functional performance. Such tools now offer the ability to perform numerous what-if type analyses to help virtually-evaluate scenarios, thereby providing enormous cost- and time-savings. However, there exist significant differences in the capabilities and ease-of-use between these tools, necessitating a careful evaluation. Hence, in this paper, we perform comparative analysis of motion data from two alternate human motion-capture systems (Vicon vs Kinect) processed using state-of-the-art computational-analysis systems (AnyBody Modeling System/Visual3D). The quantitative evaluation of a clinically relevant task (squatting) facilitates an objective evaluation of functional performance including effects of motion capture fidelity (from various sources) and the role of preand post-processing (calibration, latent dynamics estimation). Keywords: Human Modeling, Motion Capture, Kinect, Validation, Clinical Application.

1. Introduction Biomechanics research and clinical rehabilitation have long relied on quantitative motion capture coupled with subsequent computational analyses to help with diagnosis and treatment of various movement disorders (Winter 2004). The advent of the Kinect motion capture system potentially offers an opportunity to track human motion in real-time at a fraction of the cost of conventionally-equipped gait labs. Together with computational human modeling and analysis tools, this offers an opportunity to potentially gain insight into functional performance of humans outside the limiting motion-capture- lab setting. However, multiple factors need to be considered in supporting the deployment of such a framework. However, there exist significant differences in the capabilities and ease-of-use between these tools, necessitating a careful evaluation. Hence, in this paper, we perform comparative analysis of motion data from two alternate human motion-capture systems (Vicon vs Kinect) processed using state-of-the-art computational-analysis systems in the form of the AnyBody Modeling System (AnyBodyTech 2012) and Visual-3D (C-Motion 2012). In order to retain clinical relevance, we focus our attention on quantitative evaluation of squatting. Work-related musculoskeletal disorders

Email: [email protected]

are one of the greatest occupational health concerns and stooping/squatting posture is the dominant cause of the disorder. A squatting posture can be described as a “bending of the knee so that buttocks rest on or near the heels” or defined as “low working height”, “near ground level” and “at or below knee height”. The squatting posture is one of most prevalent in agriculture, construction, mining and other profession not usually considered to be physically demanding so it is important to understand and recognize the entailed risks (Fathallah and Meyers 2004).

(a) (b) Figure 1: (a) FMS scoring criteria for deep squat scoring 3 (Cook and Burton 2011), (b) Knee flexion angles for the preoperative and postoperative and control groups (Ramsey et al. 2007)

The Functional Movement Screening (FMS) test is a method of evaluating basic movement abilities. The test is comprised of seven fundamental movement patterns including deep squatting that require a balance of mobility and

1

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

stability. Deep squatting is used to assess bilateral, symmetrical and functional mobility of the hips, knees and ankles (Cook and Burton 2011). From a clinical perspective, measuring knee joint angle is a fundamental and critical technique both for diagnosis of knee deficit and monitoring improvement for treatment. Ramsey et al. (2007) introduced the effect of anatomic realignment on muscle function. They show that static alignment improves medial laxity, stability and adduction moment but requires accurate measurement of knee angle. Figure 1-(b) reproduces knee flexion angles for the pre-operative, post-operative and control groups, originally presented in Ramsey et al. (2007). Hence in this work, we focus on quantitatively comparing the knee flexion angle estimates for subjects (acquired using multiple synchronized motion capture methods and postprocessed with different biomechanical analysis software). Dutta (2012) assessed the accuracy of Kinect system by using Vicon system as a gold standard. The Kinect system can detect 3-D coordinates of markers with RMS error of 5.7mm to 10.9mm over the range of 1.0m to 3.0m. Gabel et al. (2012) examined the capability of Kinect system to extract gait information. Pedro and Caurin (2012) showed that Kinect system has good repeatability in its central region and it progressively worsens with increasing depth. Sinthanayothin et al. (2012) surveyed different techniques for vision-based human motion captures and analysis along with the capability of Kinect system for human skeleton tracking. Rincón et al. (2011) studied the feasibility of recovering human pose with data from a single camera by using particle- or Kalman-filters. However, most previous work focused on directly comparing motion-capture performance between the Kinect system and high-end measurement systems (Vicon); improving motion-capture capabilities without considering objective performance assessments; and especially without clear consideration of clinical aspects. Computational musculoskeletal analysis tools use motion-capture as input and permit estimation of latent/internal human variables (eg. muscle/tendon-/joint-reactions, mechanical work, or metabolic power consumption). Examples include both commercial tools such as Visual-3D (CMotion 2012) and AnyBody Motion System (AnyBodyTech 2012), as well as the more recent open source tools such as OpenSim (Simtk.org 2012). Musculoskeletal models are built upon a framework of constrained articulated multibody systems – with rigid skeletal bones overlaid with multiple muscles that serve to both constrain and actuate the system. The governing equations can be obtained as the constrained dynamic equations of this articulated multibody system. However, the significant actuation redundancy creates

indeterminacy for resolving muscle–actuator force distribution via inverse dynamic analyses, and is resolved by use of optimization approaches. 2. Materials and Methods Our study beings with motion-capture of squatting using two alternate synchronized systems – a high-fidelity Vicon MX System (Vicon 2013) together with a low-cost commercial-off-the-shelf Kinect sensor. The Vicon motion capture system uses high-speed (120Hz) digital cameras to track retro-reflective markers placed over body segments (head–neck, trunk, pelvis, arms, forearms, thighs and feet), from which 3D human movement is inferred using reconstructed 3D maker trajectories. We employ an 8 camera Vicon MX system that is synchronized with Kinect system via a video synchronizer (Kistler 5610), as seen in Figure 4. The Kinect system consists of Kinect sensor interfaced via either the Kinect Windows API or the Kinect Windows SDK. The Kinect system can stream color image, depth image data and also can recognize and track human skeleton joints. The Kinect system can recognize up to six human objects in the field of view and up to two objects can be tracked in detail. Furthermore it provides the information of absolute positions of joints in Kinect camera coordinate and the orientation of bone in the form of quaternions and rotation matrices.

Figure 2: Overall procedure overview: motion capture, inverse kinematic analysis, filtering/calibration of the data.

Both human motion-capture systems can provide positions of markers or anatomical landmarks. However, data from Kinect system has significant residual error and needs to be calibrated and filtered with either a kinematic-calibration or Kalman-filtering process (as illustrated in Figure 2). Additonally the position and orientation information of joints and links of human body segment are analyzed using AMS/Visual-3D. In this paper, we will consider the result from Vicon motion-capture combined with the AMS/Visual-D as gold standard – although the error from softtissue artifact needs to be considered (Benoit et al. 2006).

2

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

(a)

(a)

(b)

Figure 3: (a) Subject with reflective markers; and (b) Marker set for Vicon system

Three healthy male subjects without knee injury history were studied and marker set for AMS full body gait model are adapted. Reflective markers were fixed on 41 anatomical landmarks as presented in Figure 3 and subjects were asked to perform a series of gentle squatting trials and each subject was given several practice trial before recording. Five measurement trials were recorded for each movement task as well as a standing reference trial before the movement trials.

(b)

(c)

(d)

Figure 5: Kinect system’s skeletal frame and knee joint angle data when (a) Kinect system detects second subject, (b) and (c) part of body is not in the workspace, and (d) subject is located in ideal position

(a)

(b)

(c)

Figure 4: (a) Motion capture system configuration; Inverse Kinematic/Dynamic Analysis from (b) AMS, (c) Visual 3D

(a)

3. Results 3.1 Optimal Environment for Kinect system Kinect system’s skeletal tracking is designed to recognize users facing the sensor and Kinect system can recognize people located between 0.8 meter and 4.0 meter away, suggesting a practical range of 1.2 to 3.5 meter (Microsoft 2013). We have observed that the performance of Kinect system depends critically on the location of subject within the workspace. As shown as Figure 5, detection of second subject causes the tracking performance of the primary subject to deteriorate. Thus, although it is able to simultaneously track multiple human subjects: (i) we will restrict its use toone subject at a time; while (ii) ensuring all body segments are within the workspace of Kinect system. Figure 5(a) shows the effect of second subject detection, (b) and (c) shows the relationship between subject’s position in the workspace and performance of Kinect system, while (d) illustrates the ideal location of subject for motion capture and corresponding inferred knee joint angles.

(b)

(c)

Figure 6: Kinect system’s skeletal frame and knee joint angle data when Kinect system is located at: (a) front view, (b) side view, and (c) diagonal view

3

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

Figure 6 depicts the skeletal frames and knee joint angle captured by Kinect system during level gait. Only full-frontal configurations (where the subject is facing the sensor) provides a reasonable knee joint angle profile, and can also be observed using the gold-standard system. 3.2 Primary Inverse Kinematic Data The error of measurement system can be classified into two categories – accuracy and precision. Accuracy is the difference between the part’s measured and actual value and can be assessed in terms of bias linearity and stability. The precision is the variation when the same part is measured repeatedly with the same device and can be characterized in terms of repeatability and reproducibility. Figure 7 depicts the knee joint angle of four independent trials. Since identical squatting motions (exactly same as previous trials) cannot be realized, classical methods to assess repeatability cannot be directly employed.

magnitude than the ones from AMS and Visual 3D.

(a)

(b)

Figure 8: Joint angle plots captured by Kinect system, Vicon system and analyzed by AMS, Visual3D for (a) left knee and (b) right knee angle (a)

(b)

The knee joint angle is often extracted and used for comparisons of in gait and posture study. The accuracy of Kinect system was calculated as the difference between knee joint angle of gold standard and Kinect system. The average error and standard deviation between them are presented in Figure 9 and Table 1. The solid blue line in Figure 9 is the absolute error values of knee joint angle derived from AMS and Visual-3D which is very close to each other. The dotted red line is the error between data from AMS and Kinect system which shows significant difference in both average and STD of error.

(c)

Figure 7: Right knee joint angle from four trials of squatting captured by (a) Kinect system, (b) Vicon system/AMS, (c) Vicon system/Visual-3D

AMS and Visual-3D used same motion capture data from Vicon system and analyze it with their own human model and kinematic optimization algorithms. The knee joint angles calculated by different tool show almost same profile however data from Kinect system shows a much smaller

Figure 9: Error of joint angle (Original)

Table 1:Average and STD of the error (Original)

4

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

3.3 Kinematic Calibration

(a)

are gold standard. Before the kinematic calibration, offset of joint positions (between AMS and Kinect system’s model) needs to be applied. The least squares method offers one means for optimal estimation of the unknown parameters, (̂) that minimizes the sum square of residual errors:

where e is residual error. Given a process model: ̃

(b)

Figure 10: (a) Skeletal model of right leg; and (b) Kinematic calibration procedure

As seen in Figure 11, the joint angle shows consistent pattern comparing to the length of limb segments. The maximum deviations of the length of shank and thigh are 122mm and 190mm respectively although the lengths of segments from AMS and Visual-3D analysis are constant. Broadly we note that the Kinect system has better capability to measure the orientations of segments than measure joint positions.

where ̃ is measured value which is the position of knee and hip joint, H is basis function matrix which represent forward kinematics of two links serial chain, is true value which is the length of shank and thigh, and is measurement error (assumed as zero in this problem). The relationship between joint positions and link lengths can be written as:

[ [

]

[

]

]

[

]

where Oi are offset values of knee and hip joint position between Kinect system and AMS model which can be measured from subjects directly during motion capturing and can be considered as constant. We assumed residual error and offset as constant. Thus,

(a)

[ [

]

[

]

[

]

]

̃ (b)

The optimal estimate ̂ can be expressed as: ̂

Figure 11: (a) Length of shank and thigh; (b) Angle of right knee joint captured by Kinect system

The knee joint angle calculated from Kinect system follows the ground truth data from AMS and Visual-3D. As a first trial, we assumed the orientation information of segments captured by Kinect system is reasonably accurate, but that the length of lower limb segments are not. We will also assume that the joint positions analyzed by AMS

̂ [̂

]

(

)

̃

Figure 12 depicts the joint angle of knee joint, as calculated by forward kinematics of two link serial chain, with optimal link lengths and calibrated knee joint angle (Crassidis and Junkins 2004). The calibrated knee angles are much closer to the ground truth values (shown as dotted red line in Figure 12) as compared to the output knee angles from Kinect system. However, this requires kinematic calibration parameters (link lengths, offset values) to be additionally estimated.

5

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

(a)

Figure 13: Error of joint angles (Calibrated)

(b)

Table 3:Average/STD of the error (Calibrated) Figure 12: Joint angle plots captured by Kinect system/AMS and a kinematically-calibrated Kinect data for (a) left knee and (b) right knee

Table 2 shows the lengths of body segments captured from AMS, Kinect system and estimated by least square method. As discussed in the section above, the lengths from Kinect system (during squatting motion) are not consistent comparing with the data from AMS. However the lengths captured at initial stationary posture are very close to the data from AMS and optimal estimated values. Hence link length estimate from Kinect system (at stationary postures ) can be used to bootstrap the kinematic calibration process. The offset value between AMS and Kinect model can be derived from statistical search for the relationship between two models.

Table 2: Length of shank and thigh captured by Kinect/AMS system vs. least squares estimates

The effect of kinematic calibration is worthy of close attention, considering the simplicity of implementation. The average and STD between knee joint angle from Kinect/AMS and the calibrated Kinect system is very close to the ones from AMS and Visual-3D as shown in Figure 13 and Table 3.

3.3 Kalman Filtering As seen in the previous section, the skeleton data obtained from Kinect system is not accurate. The knee angle calculated from the skeleton points shows large discrepancy from the ground truth (AnyBody result based on Vicon data). However, qualitatively it still captures the trend, although significant quantitative difference remain. Therefore we sought to improve the performance by Kalman Filtering. While the previous attempt of calibration shows good result, it relies on accurate Vicon data for each trial. The goal herein is to use only Kinect to obtain an estimate as close as possible to the Vicon system. Based on observed data, we found that the initial skeletal data from standing posture is relatively good; the link lengths computed are very close to ground truth. As the subject starts moving, the data gets worse, but one is able to discern the movement pattern. Therefore, we use the following assumptions: (a) subject’s leg length is constant allowing capture of the best data from standing posture, then we just use the initial leg length; (b) since Kinect is still able to capture rough movement pattern, we run a first pass of calculating the knee angle from raw kinect data, and use that to drive the motion model; (c) since the main component of the motion is planar, as a starting point and for simplicity, we use the planar two link model for each leg. We assumed that the ankle position ( ), is fixed during squatting motion. We set the origin of coordinate frame at the ankle. For the knee ( ), the forward kinematics is ( ) , differentiating to get ̇ ( ) ̇ , and ̈ ̈ ̇( ) ̇ ( ) . Similarly for the hip

6

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

(

), ̈

and

) ̇

̇(

( ) , ̇

̇ ( ̇

Setting the process model is: ̇

[

]

̇ ̇

[ ̇

̇

) ̈

) ̇ ̈

̇

̇

(

̇

̈

] ̈

̇

̇

,

. ̇

̇ ,

,

̈

Figure 14 and Table 4 show the estimation results. To illustrate the effectiveness, the noise covariance value is tuned according to only trial RB111 left leg (first trial), then the same value is applied to the rest of the trials. As can be seen, the estimated angles are consistently and significantly improved from the raw Kinect data, for all other cases. We anticipate getting better estimates when we have more trial data and obtain a better statistical noise characteristic.

and the observation model is the skeletal points at the ankle, knee and hip: ,

where are the corresponding process and observation noise, which are assumed to be Gaussian, with covariances . Then we can apply the Kalman Filter (Kalman 1960) to try get an estimate of the joint positions, from which we calculate knee angle: Figure 15: Error of joint angles (Kalman Filter)

Predict: ̂

̂

Update: ̂

̂

( (

̂

)

)

where the Kalman gain is computed as: (C

)

(a)

(b)

Figure 14: Kalman filter result of knee angles from three trials using same parameter

Table 4: Average/STD of the error (Kalman Filter)

4. Discussion Two motion capture systems – a low-fidelity and low-cost Kinect framework and the moreexpensive, higher fidelity Vicon – were examined to aid quantitative knee-angle estimation in a clinically-focused squatting study. In particular, we sought to compare the systems in terms of quantitative performance criteria including workspace, sampling rate, accuracy and portability. The combination of Vicon Motion-capture with AMS/Visual-3D post-processing yields outstanding performance (with huge workspace, high resolution and sampling rate) but with limited portability and high cost. The commercial-off-the-shelf Kinect system offers an ultra-mobile (2.2 kg) solution costing less than $100 but has small workspace, relatively low resolution. Thus direct application of the Kinect system to clinical or research work (without post-processing of raw data) tends to be limited. However, we noted that with suitable post processing offers potential for applicability for clinically relevant use. Specifically, the two methods (kinematic calibration and Kalman filtering) analyzed for the squatting study showed good performance versus

7

Seung-kook Jun, A Comparative Study of Human Motion Capture and Analysis Tools

the clinical gold standard (Vicon and AMS/Visual3D). Further studies are currently underway to statistically validate these results. Acknowledgements This work was supported by the National Science Foundation Award CNS-1135660. The authors would like to thank Robin Herin for his help with the data collection. References AnyBodyTech 2012. AnyBody Modeling System Tutorial. Version 5.2.0. Benoit, D., D. Ramsey, M. Lamontagne, L. Xu, P. Wretenberg and P. Renstroma 2006. "Effect of skin movement artifact on knee kinematics during gait and cutting motions measured in vivo." Gait & Posture 24(2): 152-164. C-Motion. 2012. "Visual-3D." from http://www.cmotion.com/products/visual3d/. Cook, G. and L. Burton 2011. Movement, The Book, On Target Publications. Crassidis, J. L. and J. L. Junkins 2004. Optimal Estimation of Dynamic Systems, CRC PRESS.

Ramsey, D., L. Snyder-Mackler, M. Lewek, W. Newcomb and K. Rudolph 2007. "Effect of Anatomic Realignment on Muscle Function During Gait in Patients With Medial Compartment Knee Osteoarthritis." Arthritis & Rheumatism 57(3): 389–397. Rincón, J. M. d., D. Makris, C. O. Uruñuela and J.C. Nebel 2011. "Tracking Human Position and Lower Body Parts Using Kalman and Particle Filters Constrained by Human Biomechanics." IEEE Transactions on Systems, Man, and Cybernetics 41(1). Simtk.org. 2012. "OpenSim ", from https://simtk.org/project/xml/downloads.xml?group _id=91. Sinthanayothin, C., N. Wongwaen and W. Bholsithi 2012. "Skeleton Tracking using Kinect Sensor & Displaying in 3D Virtual Scene." International Journal of Advancements in Computing Technology(IJACT) 4(11.23). Vicon 2013. "Vicon MX System." Winter, D. A. 2004. Biomechanics and motor control of human movement, John wiley & sons Inc.

Dutta, T. 2012. "Evaluation of the Kinect sensor for 3-D kinematic measurement in the workplace." Applied Ergonomics 43: 645-649. Fathallah, F. A. and J. M. Meyers 2004. Stooped Squatting Postures Workplace. http://agergo.ucdavis.edu/papers/435028.Stooped_smallsize .pdf. Gabel, M., R. Gilad-Bachrach, E. Renshaw and A. Schuster 2012. Full Body Gait Analysis with Kinect. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. San Diego. Kalman, R. E. 1960. A New Approach to Linear Filtering and Prediction Problems. Microsoft. 2013. "Kinect skeletal tracking." from http://msdn.microsoft.com/enus/library/hh973074.aspx. Pedro, L. and G. Caurin 2012. Kinect Evaluation for Human Body Movement Analysis. The Fourth IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics. Roma, Italy. June 24-27, 2012.

8

A Comparative Study of Human Motion Capture and ...

analysis tools that can provide significant insights into the functional performance. Such tools now ... and clinical rehabilitation have long relied on quantitative.

1MB Sizes 2 Downloads 255 Views

Recommend Documents

A STUDY OF Comparative anatomy of papillary muscles of human ...
A STUDY OF Comparative anatomy of papillary muscles of human, sheep, cow and pig.pdf. A STUDY OF Comparative anatomy of papillary muscles of human, ...

A COMPARATIVE STUDY OF NURSING EDUCATIONAL SYSTEM ...
Retrying... Whoops! There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. Main menu. Whoops! There was a problem previewing A COMPARATIVE STUDY OF NURSING EDUCATI

A COMPARATIVE STUDY OF DISCRIMINATIVE ...
Center for Signal and Image Processing, Georgia Institute of Technology. 75 Fifth ... we call cross-layer acoustic modeling in that the model discrimina- tion is often at ..... lated cross-layer error cost embedded on the WSJ0 LVCSR database.

Design and Implementation of e-AODV: A Comparative Study ... - IJRIT
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, ... Keywords: Wireless mobile ad hoc networks, AODV routing protocol, energy ... In order to maximize the network life time, the cost function defined in [9] ...

A comparative study on engine performance and emissions of ...
Page 1 of 7. Indian Journal of Engineering & Materials Sciences. Vol. 21, August 2014, pp. 438-444. A comparative study on engine performance and emissions of biodiesel and JP-8. aviation fuel in a direct injection diesel engine. Hasan Yamika. , Hami

Design and Implementation of e-AODV: A Comparative Study ... - IJRIT
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, ... In order to maximize the network life time, the cost function defined in [9] ...

A comparative study of ranking methods, similarity measures and ...
new ranking method for IT2 FSs and compares it with Mitchell's method. Section 4 ... edge of the authors, only one method on ranking IT2 FSs has been published, namely Mitchell's method in [24]. ...... The authors would like to thank Professor David

pdf-1828\sufism-and-taoism-a-comparative-study-of-key ...
Try one of the apps below to open or edit this item. pdf-1828\sufism-and-taoism-a-comparative-study-of-key-philosophical-concepts-by-toshihiko-izutsu.pdf.

Motion Capture and Its Application for Vehicle Ingress/Egress
identify the root cause since the vehicle package ... seating bucks with targeted vehicle package ... built for various packaging and ergonomics analysis.

Motion Capture and Its Application for Vehicle Ingress/Egress
identify the root cause since the vehicle package ... built for various packaging and ergonomics analysis ... that are actuated by software-multiplexed electric.

A comparative study of ranking methods, similarity ...
An illustration of eA 6 eB is shown in Fig. 6. The following ...... 0. 9. Low amount .31 .31 .36 .37 .40 .40 .99 .99 .99 .81 .50 .50 .38 .29 .29 .15 .15 .15 .02 .02 .02. 0.

A comparative study of different feature sets for ...
On experimentation with a database of. 3000 samples, the .... respectively. Once a feature set is fixed up, it is left with the design of a mapping (δ) as follows:.

A comparative study of probability estimation methods ...
It should be noted that ζ1 is not a physical distance between p and ˆp in the .... PDF is peaked or flat relative to a normal distribution ..... long tail region. .... rate. In the simulation, four random parameters were con- sidered and listed in

A Comparative Study of Differential Evolution, Particle Swarm ...
BiRC - Bioinformatics Research Center. University of Aarhus, Ny .... arPSO was shown to be more robust than the basic PSO on problems with many optima [9].

A comparative study on Indium Phosphide and α ...
state transmitters in tracking radars, missile seekers, radi- ometers, and mmwave .... sentially depends on the transit time of charge carriers to cross the depletion ...

A Comparative Study of Different Presentation ...
Doctor Aiguader 80, 08003 Barcelona, Spain. david.andreu@ upf.edu. ..... Blocking of free binding sites was performed ... ducted using Microsoft Excel software.

A comparative study of different feature sets for ...
8. (b). Fig 6. An illustration for computation of the row wise longest–run feature. .... [1] A. Amin, “Off-line Arabic Character Recognition: the State of the Art”, Pattern ...

Comparative Study of Reversible Image ...
Hiren R. Soni , IJRIT. 161. IJRIT International Journal of Research in Information Technology, Volume 1, Issue 4, April 2013, Pg. 31-37. International Journal of ...

2. Treaty Making Power Problems: A Comparative Study and ...
ASEAN Sectorial Mutual Recognition Arrangements (MRAs), ASEAN- Australia-New Zealand FTA, and ASEAN-India FTA. 6 Similarly, the. ASEAN Charter and ...

A Comparative Study of Test Data Dimensionality ...
Although both have different points of departure, the essentially and strictly unidimensional IRT models both imply weak LI. For analyzing empirical data, both ...

A Comparative Study of Low-Power Techniques for ...
TCAM arrays for ternary data storage, (ii) peripheral circuitry for READ, WRITE, and ... These issues drive the need of innovative design techniques for manufacturing ..... Cypress Semiconductor Corporation, Oct. 27, 2004, [Online], Available:.

A Comparative Study of Anomaly Detection Techniques ...
approach with anomaly detection techniques that have been proposed earlier for host/network-based intrusion detection systems. This study enables gaining further insights into the problem of automatic detection of web defacements. We want to ascertai

A Comparative Study of Methods for Transductive ...
beled data from the target domain are available at training. We describe some current state-of-the-art inductive and transductive approaches and then adapt ...