Feature Tracking Using Particle Filter In Rope Skipping .
International Journal of Computer Applications (0975 – 8887)International Conference on Cognitive Knowledge Engineering 2016Feature Tracking using Particle Filter in Rope Skippingfor Gross Motor Skill DevelopmentStephen KarungaruKenji MatsuuraNada GotodaGraduate school of Scienceand Technology,University of Tokushima,Tokushima, JapanGraduate school of Scienceand Technology,University of Tokushima,Tokushima, JapanInformation Technology Center,Kagawa University,Takamatsu, JapanABSTRACTLearning a new skill for physical development can be adaunting task for many novice persons. To support suchlearners, an intelligent system is required to guide them in thelearning process. In this paper, as first part of such a system,we propose a feature detection and tracking algorithm that canbe used during rope skipping skill development using colorprocessing and the particle filter. The data used is capturedusing a camera placed on the side of the learner. The learnerwears markers on the head, hands and ankles; a marker is alsoattached on the rope to capture rope rotation. Initial pointdetection is achieved using HSV color space thresholding.The particle filter is then used to track these featuresespecially because of misdetections due to noise and blurringdue to rope speed. In this work, the rope skill attempted is thelearning to do the “double under” jump. A “double under”jump is defined as completing two rope rotations per jump.Experimental results prove that this is an effective method foraccurate feature detection and tracking.General TermsPattern Recognition, Object tracking, Image processing,Gross Motor skillsKeywordsMotor skill, particle filter, learning support, Curve fitting1. INTRODUCTIONRecently, the use of technology to support humans learn anddevelop motors skills has become paramount. The technologycan be used to monitor and analyze the learner’s process andoffer advice as required to aid faster and easier acquisition ofthe skill. Moreover, learners can follow their progress usingdata from such systems including visual and audioinformation [1]. Motor skills can be acquired through specifictraining which means that it is not innate ability but potentialchange against specific conditions [2]. In this work, out targetto develop a support system for rope skipping.Rope skipping is a simple, fun and easy-to-learnactivity that is great for fitness. All that is required is arope and some little space to play. Rope skipping involvesone or more participants who jump over a rope swung so thatit passes under their feet and over their heads [3]. Ropeskipping as a sport including definitions and howtos isexplained in details in [7]. Three main variations exists, but inthis work we are interested in the most basic one that involvesa single participant rotating and jumping over therope. Learning this kind of rope skipping is easy as itinvolves just one rope rotation per jump. In our work, wehope to design a support system for an advanced ropeskipping variation referred to as “double under”. Doubleunder involves attaining two rope rotations per jump. Thisskill is on a different level to the single jump and isusually very difficult to master because of the rope speedvariations, jumping height, rhythm and technique. Wehope that our work will make it easy for learners toacquire this skill. As a first step, we will concentrate onfeature detection from video to facilitate the developmentof the required support system.In a related work, Yoshioka et. al. [3] used imageprocessing to analyze a monitored video and give theappropriate feedback to the learners [4]. They use imageprocessing technology on a desktop computer, which meansthat analysis can be performed using an efficient processorand the results to improve a player can be delivered in arelative larger display at asynchronous timing against theperformance. However, the continuous detection and trackingof the markers was not accurate enough because of imagecapture conditions and occlusions.Ideally, simple programing to capture a motor skillbasic movements is possible if skill conditions and otherenvironment conditions are known and fixed. This is referredto as a closed skill an example of which is rope skipping. Wecapture a video of a person performing rope skipping and useit to support the learning process. Rope skipping is a repetitiveprocess, so as an initial we hope to accurately detect and trackthe heel, rope, hand and head positions using the particlefilter. The use of the particle filter is necessary because simpledetections using color information cannot be obtained steadilydue to noise, capture position, rope speed, etc. The particlefilter methodology is used to solve Hidden Markov Chain(HMM) and nonlinear filtering problems arising in signalprocessing and Bayesian statistical inference [5].This work aims at accurately detecting the body partsheavily involved in the skill development including the hands,feet and head. The rope position will also be tracked at alltimes. Moreover, we will calculate the rope speed andpositions during different parts of the jump. The motivation isbased on the schema theory that states that as we learn amotor skill, we develop a rule that shows the relationshipbetween movement outcomes and the intended goal, theconditions of the performance setting, and the details of themotor program created to control the movement [6].Capturing the rope speed, angle, jump height, etc. can help asdevelop rules required for double under jump.The rest of this paper is organized as follows. Section 2discusses the marker detection from video using thresholds setin the HSI color space. Marker tracking using the particlefilter is also discussed. Feature extraction is presented insection 3 with the experiments and results in section 4. Finallyare the discussions and conclusion sections.25
International Journal of Computer Applications (0975 – 8887)International Conference on Cognitive Knowledge Engineering 20162. MARKER DETECTION2.1 Set upThe data used in this work is captured outdoors. Thebackground is a solid wall of uniform color and patterns foreasy and fast background extraction. The camera is placedabout 3 meters on the side of the rope skipping subject. Thesubject wears the following color markers, fig. 1. Thesemarkers are necessary to track rope and body parts movementduring the exercise. Rope: Feet: Hands: Head:RedGreenYellowBlueFig 2: Marker positionsTo solve these problems we introduce the particle filterto estimate the position of the markers at all times andconditions. The positions of the markers are captured in everyframe of the video. This data is used to extract the requiredfeatures.2.3 Marker Tracking: Particle FilterFig 1: Marker positions2.2 Marker DetectionAfter video capture the first process applied is backgroundsubtraction to extract the person location. The clips arecaptured on a relatively simple background to make thisprocess simple.Each marker is then detected using the HSI color space byassigning the appropriate thresholds based on the samplecolors extracted from the captured images. The thresholdsassigned are as follows, (minimum and maximum). Rope:Feet:Hands:Head:RedGreenYellowBlueThe filtering problem consists of estimating the internal statesin dynamical systems when partial observations are made, andrandom perturbations are present in the sensors as well as inthe dynamical system. The objective is to compute theconditional probability (posterior distributions) of the states ofsome Markov process, given some noisy and partialobservations [3]. In Markov models, given the current state,the past and future states are conditionally independent.Therefore, only the current state is required.In non-linear and non-gaussian problems the particlefilter cam be applied. That is based on a hypothesis, it canapproximates the posterior distribution by a group of weightedparticles. Weights are assigned to the particles based on alikelihood score and re-sampled according to a given model.In this work we must track multiple feature pointsincluding the head, hand, feet and rope. The images are firstprocessed using the HSI color space to extract the locations ofthe objects to be tracked. Every object is tracked individually.The number of particles per object is set to 50 based onexperiments. Each particle will represent a state the objectmay be in at a given time. Different colors are used todifferentiate between the markers being tracked, fig 3.(130,80,90) (50,150,250)(60,80,90) (80,200,200)(20,50,150) (50,150,250)(100,170,100) (180,250,250)These thresholds were determined using experimentation.A more intuitive method for marker detection is planned for inthe future.These simple threshold can detect the marker positionsfast and at fairly good accuracies (85%). However, not allmarkers can be detected in all frames because of image noiseand occlusion. Figure 2 shows the spikes (black oval regions)created on the jump curve due to the mis-detections for therope. The ground position is at 100 on the Y-axis.Fig 3: Particle Filter initial locations26
International Journal of Computer Applications (0975 – 8887)International Conference on Cognitive Knowledge Engineering 20163. FEATURE EXTRACTIONTo support learners in rope jumping, several features that canbe used to define the process that are necessary. Theseinclude: Rope rotation speedJump heightHead and feet positionsHand movementsBody form during jumpRhythm3.1.1 Process Start pointThe feature capture starting position of this experiment iswhen the rope is just above the head. At this position, the headand feet are approximately on a straight vertical line runningfrom the head to the feet. The hands are in slightly bentposition in front of other points (A bit forward). Moreover, allthe markers (except the rope which is at its maximum point)are at their lowest point, Fig 4, 5.Fig 6: Successful “double under”However, during “double under”, the horizontalmovement of the hands is vital. The hands move in a stretched“8” movement for some subjects as shown in fig. 7, 8 (a).However, for some subjects the hand movements are asshown in fig. 8(b).Fig 7: Successful “double under”: marker movementsFig 4: Starting Point: The curves from top shows the head,hand, feet and rope positions.(a)Fig 5: Starting Point(b)Fig 8: Hand movements during a Successful “doubleunder” for 2 subjects3.1.2 Double under3.1.3 Rope SpeedA successful double under jump will be defined as the ropepassing the starting point twice before the feet again touchesthe floor, Fig 6.In most cases (people), the rope, head and to someextent, the feet positions can be well represented using onlythe vertical positions of the markers.To successfully complete double under, the rope speed varieswidely during the two full rotations. Generally, the speed isslowest above the head and under the feet on every rotation.The speed increases when the rope is behind the subject andmaximizes when the rope is just in front of the subject duringthe start of the second rotation.27
International Journal of Computer Applications (0975 – 8887)International Conference on Cognitive Knowledge Engineering 2016Table 1: Rope Feature Results4. EXPERIMENTS AND RESULTSTo prove the effectiveness of the proposed method,experiments were conducted. The main aim of theexperiments is to accurately detect the rope, head, hands andfeet markers using color information and particle filter, andextract features that can be used to create a support system fordouble under learners. All the volunteer subjects in the workcan execute the double under jump.4.1 Experimental dataThe data used in this experiment was captured outdoors nextto a building. The capture background is relatively uniformand easy to extract.There are eight subjects and a total of 10 video clipseach about 11 seconds long (About 380 frames at 30fps). Allthe subjects can perform the double under jump and are askedto do it a minimum of 3 times per session. Some of the clipshave low contrast making it harder to detect the markersaccurately.This work was carried out using a computer with anIntel Core(TM) i7-4790 4GHz CPU.Average Rope Rotation (sec)RopeSpeedFirst 24644.2 ResultsThe particle filter performed as expected by tracking themarker position especially when the color information failedto extract it position. In this work, 50 particles per markerwere sufficient because the initial marker positions could beestimated fairly well.Different colored marker we used to simultaneouslytrack all the markers. Figure 9 shows the plot of the ropeheight and time taken to complete the jump for all thesubjects. The green line shows the slowest and the orange onethe fastest subject.Another observation was in the rope position whenmaximum or above average speed was observed. In all cases,the maximum speed was attained during the second rotation,with the rope position directly above the subject as shown infig. 10. At this point, the feet position are is also at the highestpoint. The rope passes just under the feet to complete thesecond jump and therefore the double under.Fig 10: Highest rope speed regions: In general the speed ishigher in front of the subject (red triangle)4.3 DiscussionsFig 9: Rope height during the double under jumpTable 1 below shows the results of the rope featuresextracted by our proposed method. The features include theaverage time taken for first and second jump and the max ropespeed for each subject. The calculations are based on a framerate of 30.The head, hand and feet vertical movement is consistentcovering the period required to complete two rope rotationsregardless of the rope speed. Some variations were observedin horizontal hand movements which depend on the roperotating style of the subject.From the data and results, we can conclude that the doubleunder jump can be accomplished by combination of high ropespeed and medium jump or medium rope speed and highjumps. Jumping height and rope rotation speed also seem todepend on learner. Therefore, because of the this individuality,either each learner might require a more specialized supportsystem or the system should be modelled to account for thesedifferences.The average double under jump takes about 2.5 secondsfor most subjects. However, subject 3 still accomplished thedouble under jump in spite of lower than maximum ropespeed of 44 (above 50 for other subjects) and a duration ofabout 1.5 second slower. Subjects 7 and 8 achieved fastertimes than all other subjects.28
International Journal of Computer Applications (0975 – 8887)International Conference on Cognitive Knowledge Engineering 20165. CONCUSION AND FUTURE WORKS7. REFERENCESIn this work, as a first step in the creation of an intelligentrope skipping support system, we proposed a feature detectionand tracking algorithm using particle filter that can accuratelyextract the markers required by the system. Initial pointdetection is achieved using HSV color space thresholding.The particle filter is then used to track these featuresespecially because of misdetections due to noise and blurringdue to rope speed. In this work, the rope skill attempted is thelearning to do the “double under”. Experimental results provethat this is an effective method for accurate feature detectionand tracking.Consistent results were observed in all subjects. Theaverage jump duration was about 2.6 seconds. The fastest was2.21 seconds and the slowest at 4.13 seconds. However, thejumping rhythm (features timings) was similar for all thesubjects.In future the features detected and tracked in this workwill be incorporated in the support system to learn rulesrequired to effectively aid rope skipping learners. Moreover,to improve the system the individuality of each learner needsto be established for easier support.[1] Clark J.M. and Paivio A. Dual coding theory andeducation. Educational Psychology Review, 3(3):149210, 1991[2] Richard A. Schmidt. Motor Learning & Performance:From Principles to practice. Human Kinetics Books,1991.[3] Yoshioka S. Yamada K. and Matsuura K. Supportingsystem for the form improvement on rope skipping skillby image processing. In 2014 IIAI InternationalConference on Advanced Applied Informatics, pp.328331, 2014.[4] Yoshioka S. Matsuura K. and Gotoda N. Hand-motionanalysis for development of double-unders skill. In 9thInternational Conference in Knowledge Based andIntelligent Information and Engineering Systems, pages775-783, 2015.[5] Particle filter,https://en.wikipedia.org/wiki/Particle filter (last modifiedon 21 October 2016)[6] Schmidt, R. A. A schema theory of discrete motor skilllearning. Psychological Review, 82, 225–260, 1975.[7] Carolyn Barker and Kym Warner: Australian RopeSkipping Association: Level 1 Coaching anualRope-Skipping.pdf.6. ACKNOWLEDGMENTSThis work was wholly supported by Japan Society for thePromotion of Science (JSPS) KAKENHI Grant NumberJP15K01072.IJCATM : www.ijcaonline.org29
Rope skipping is a simple, fun and easy-to-learn activity that is great for fitness. All that is required is a rope and some little space to play. Rope skipping involves one or more participants who jump over a rope Capturing the rope speed, angle,
Cook Medical Bird’s Nest Filter: C. Vena Tech LP FIlter: D. Bard Simon Nitinol Filter: E. Bard Recovery G2 Filter: F. Cook Medical Günther Tulip Filter: G. Cook Celect Filter: H. Cordis OPTEASE Filter: I. Argon Option Elite Filter: J. Crux VCF: K. ALN Optional Filter: Many other IVC f
Filter Gallery dialog box A. Preview B. Filter category C. Thumbnail of selected filter D. Show/Hide filter thumbnails E. Filters pop‑up menu F. Options for selected filter G. List of filter effects to apply or arrange H. Filter effect selected but not applied I. Filter effects applied cumulatively but not selected J. Hidden filter effect Display the Filter Gallery
cava filters, such as Greenfield filter, Vena Tech filter, Bird s nest filter, Simon-Nitinol filter, TrapEase filter, Günther tulip filter, Antheor filter, Neuhaus protect filter, and Recovery- Nitinol filter, have been develope
BENZ S550 W251 BENZ S550 W221 Car Model BENZ S Air filter Oil filter Fuel filter Cabin filter GA-22088U GO-22007E GO-22003E GC-22136 GA-22088U GC-22136 . 08-12 BENZ SL63 AMG 6.2 GO-22019E BENZ SL Car Model BENZ Sports Car SLK Air filter Oil filter Fuel filter Cabin filter 09-11 B
EMI FILTER There are several types of filters, for example, passive EMI filter, active EMI filter and hybrid EMI filter. However, the passive EMI filter is simpler than others. A typical procedure used in [4, 5] is considered to design power line EMI filter. A basic network topology is used, C-L-C for DM filter and L-C
Particle Properties Particle size By far the most important physical property of particulate samples is particle size. Particle size measurement is routinely carried out across a wide range of industries and is often a critical parameter in the manufacture of many products. Particle size has a direct influence on material properties such as:
q C to heat exchanger for test and evaluation Solex/VPE/Sandia particle/sCO2 shell - and - plate heat exchanger Heat duty 100 kW T particle,in 775 q C T particle,out 570 q C T sCO2,in 550 q C T sCO2,out 700 q C I 6 0.5 kg/s High - Temperature Particle Receiver Particle receiver testing at the National Solar Thermal Test Facility
property tests. The best particle model had a particle coefficient of restitution of 0.6; particle static friction of 0.45 for soybean-soybean contact (0.30 for soybean-steel interaction); particle rolling friction of 0.05; normal particle size distribution with standard deviation factor of 0.4; and particle shear modulus of 1.04 MPa. Keywords.
