首页 | 本学科首页   官方微博 | 高级检索  
相似文献
 共查询到20条相似文献，搜索用时 546 毫秒
1.
One of the most serious obstacles to accurate quantification of the underwater motion of a swimmer's body is image deformation caused by refraction. Refraction occurs at the water‐air interface plane (glass) owing to the density difference. Camera calibration‐reconstruction algorithms commonly used in aquatic research do not have the capability to correct this refraction‐induced nonlinear image deformation and produce large reconstruction errors. The aim of this paper is to provide a through review of: the nature of the refraction‐induced image deformation and its behaviour in underwater object‐space plane reconstruction; the intrinsic shortcomings of the Direct Linear Transformation (DLT) method in underwater motion analysis; experimental conditions that interact with refraction; and alternative algorithms and strategies that can be used to improve the calibration‐reconstruction accuracy. Although it is impossible to remove the refraction error completely in conventional camera calibration‐reconstruction methods, it is possible to improve the accuracy to some extent by manipulating experimental conditions or calibration frame characteristics. Alternative algorithms, such as the localized DLT and the double‐plane method are also available for error reduction. The ultimate solution for the refraction problem is to develop underwater camera calibration and reconstruction algorithms that have the capability to correct refraction  相似文献   

2.
One of the most serious obstacles to accurate quantification of the underwater motion of a swimmer's body is image deformation caused by refraction. Refraction occurs at the water-air interface plane (glass) owing to the density difference. Camera calibration-reconstruction algorithms commonly used in aquatic research do not have the capability to correct this refraction-induced nonlinear image deformation and produce large reconstruction errors. The aim of this paper is to provide a through review of: the nature of the refraction-induced image deformation and its behaviour in underwater object-space plane reconstruction; the intrinsic shortcomings of the Direct Linear Transformation (DLT) method in underwater motion analysis; experimental conditions that interact with refraction; and alternative algorithms and strategies that can be used to improve the calibration-reconstruction accuracy. Although it is impossible to remove the refraction error completely in conventional camera calibration-reconstruction methods, it is possible to improve the accuracy to some extent by manipulating experimental conditions or calibration frame characteristics. Alternative algorithms, such as the localized DLT and the double-plane method are also available for error reduction. The ultimate solution for the refraction problem is to develop underwater camera calibration and reconstruction algorithms that have the capability to correct refraction.  相似文献   

3.
One of the most serious obstacles to accurate quantification of the underwater motion of a swimmer's body is image deformation caused by refraction. Refraction occurs at the water-air interface plane (glass) owing to the density difference. Camera calibration-reconstruction algorithms commonly used in aquatic research do not have the capability to correct this refraction-induced nonlinear image deformation and produce large reconstruction errors. The aim of this paper is to provide a thorough review of: the nature of the refraction-induced image deformation and its behaviour in underwater object-space plane reconstruction; the intrinsic shortcomings of the Direct Linear Transformation (DLT) method in underwater motion analysis; experimental conditions that interact with refraction; and alternative algorithms and strategies that can be used to improve the calibration-reconstruction accuracy. Although it is impossible to remove the refraction error completely in conventional camera calibration-reconstruction methods, it is possible to improve the accuracy to some extent by manipulating experimental conditions or calibration frame characteristics. Alternative algorithms, such as the localized DLT and the double-plane method are also available for error reduction. The ultimate solution for the refraction problem is to develop underwater camera calibration and reconstruction algorithms that have the capability to correct refraction.  相似文献   

4.
水下三维物空间坐标的重构精度分析   总被引：1，自引：0，他引：1  
目的:水下运动学数据的精度对于游泳项目的技术动作分析十分重要,利用防水摄像装置对标定框架上的有效点在水下和空气中进行三维重构,旨在验证不同测试环境和不同大小的标定空间对水下三维重构精度的影响,为水下游泳的器材研发提供一定的参考和依据。方法:2台水下摄像机同步拍摄放置在泳池里的标定框架(2 m×2 m×1 m),标定框架采取2种不同大小的标定空间,分别对同样数目的控制点(12个)和有效点(4个)分别在水下和空气中进行三维重构,重构精度通过4个有效点的均方根(RMS)重构误差来表示,利用DLT算法和ARIEL软件进行标定重构和数字化。结果发现,在整体标定空间控制点的RMS重构误差在X、Y、Z轴上为:(ⅰ)5.60 mm、10.44 mm和3.74 mm(水下),(ⅱ)4.89 mm、4.89 mm和3.96 mm(空气);局部标定空间为(ⅰ)5.29 mm、9.18 mm和3.02 mm(水下)(ⅱ)1.44 mm、1.11 mm和3.25 mm(空气)。结论:由于光线折射不论标定空间的大小,水下三维重构所得到的RMS重构误差都高于空气中的测量,Y轴最为明显;随着标定空间的增大,两种测试环境下的RMS重构误差均随之增大。  相似文献   

5.
三维运动影像分析通常采用的算法是直接线性变换法(DLT),而控制点的空间三维坐标是应用直接线性变换法进行标定基本参数,它目前是通过三维标定框架的已知控制点的坐标所获得的。由于三维标定框架所覆盖的空间有限,对于较大运动范围运动项目的三维运动影像分析会造成较大的误差。研究目的是在全站仪(Topcon GPT-3002 N)为硬件的基础上,以对象编程语言C#为开发工具开发了三维影像分析控制点测量与坐标转换软件系统。此系统为对运动空间任意布置的控制点,或标定框架上的控制点的三维坐标的精确测量提供了实用的工具,也为在运动技术诊断和分析中应用三维跟踪扫描影像分析方法奠定了基础。此外,还对控制点目标观测与坐标框架换算系统的需求、总体架构、数据库设计、算法设计、具体功能和应用实现等方面进行了较为详细的论述。  相似文献   

6.
李旭鸿  侯曼 《体育科研》2005,26(5):73-75
直接线性变换(DLT)算法是利用平面图像进行三维重构的基本算法,本文从框架位置的角度对三维重构的影响进行讨论,由此得出各个框架位置下控制点的平均相对误差,结果表明框架位置对三维重构具有一定的影响,同时相机角度对三雏重构具有深远的意义。  相似文献   

7.
通过引入摄影机镜头的内、外方位元素的约束条件,导出了一种修正的DLT(MDLT)方法。该方法用于影片和录相的实测,得到一些有意义的结果。当测量的标定控制点点数较少(6～8个),且这些点的位置在空间和成像面分布不均匀时,用MDLT计算的结果比DLT的结果要好,误差可降低80％以上。  相似文献   

8.
通过水槽的水下摄像系统,对3名上海优秀运动员和4名一线运动员进行水下打腿动作测试录像,利用图像解析软件对技术录像进行分析,结果发现：优秀组队员的水下潜泳腿连贯性优于一线组队员;个性化的专项技术训练、能力训练、核心力量训练等有利于打腿能力的提高,为运动成绩的提高起到促进作用。  相似文献   

9.
Stereo camera systems have been used to track markers attached to a racket, allowing its position to be obtained in three-dimensional (3D) space. Typically, markers are manually selected on the image plane, but this can be time-consuming. A markerless system based on one stationary camera estimating 3D racket position data is desirable for research and play. The markerless method presented in this paper relies on a set of racket silhouette views in a common reference frame captured with a calibrated camera and a silhouette of a racket captured with a camera whose relative pose is outside the common reference frame. The aim of this paper is to provide validation of these single view fitting techniques to estimate the pose of a tennis racket. This includes the development of a calibration method to provide the relative pose of a stationary camera with respect to a racket. Mean static racket position was reconstructed to within ±2 mm. Computer generated camera poses and silhouette views of a full size racket model were used to demonstrate the potential of the method to estimate 3D racket position during a simplified serve scenario. From a camera distance of 14 m, 3D racket position was estimated providing a spatial accuracy of 1.9 ± 0.14 mm, similar to recent 3D video marker tracking studies of tennis.  相似文献   

10.
The use of digital camcorders in biomechanical analyses can introduce errors due to inter-camera time offsets. The faster the motion being recorded the greater the error. A sequential synchronization method was developed in this study to achieve sub-field camera synchronization for multiple camcorders through numerical optimization. A recreational golfer performed ten drives while being recorded with four digital camcorders (60-Hz sampling rate). Video signals were sampled at 10,000 Hz to determine the actual inter-camera time offsets. The optimized inter-camera time offsets were computed based on three markers placed on the shaft, each separated by 200 mm. The inter-camera time offset error was computed as the difference between the optimized and actual inter-camera time offsets. The inter-camera time offset error reduced on average from 0.518 to 0.019 fields (1 field = 16.7 ms) or less due to sequential optimization. The optimized global reconstruction errors were less than 19% of the unadjusted values. It was concluded that the ability to synchronize multiple (two or more) cameras using a sequential sub-field optimization strategy promises to extend the use of relatively inexpensive digital camcorders to motions considered too fast for the low field rates of such cameras. The sequential approach presented provides a balance between computation time and reconstruction accuracy.  相似文献   

11.
Current trends in swimming biomechanics are focused on accurate measurements. Nowadays, reliable calibration methods have been proposed to reach an accuracy of about 1 mm on rigid structure. But the question remains about the final accuracy for three-dimensional hand kinematics measurement during the underwater phase of front crawl swimming. Furthermore, most research is based on manual tracking with two or more cameras. In this paper we propose a protocol to acquire three-dimensional hand kinematics when swimming in a specific pool with a motion analysis system behind windows. Results highlight the benefits of using such a system in terms of accuracy and feasibility: the time allowed for post-processing is ten times lower and the quantified improved accuracy is better than with manual tracking.  相似文献   

12.
Cost effective, quantifiable assessment of lower extremity movement represents potential improvement over standard tools for evaluation of injury risk. Ten healthy participants completed three trials of a drop jump, overhead squat, and single leg squat task. Peak hip and knee kinematics were assessed using an 8 camera BTS Smart 7000DX motion analysis system and the Microsoft Kinect® camera system. The agreement and consistency between both uncorrected and correct Kinect kinematic variables and the BTS camera system were assessed using interclass correlations coefficients. Peak sagittal plane kinematics measured using the Microsoft Kinect® camera system explained a significant amount of variance [Rangehip = 43.5–62.8%; Rangeknee = 67.5–89.6%] in peak kinematics measured using the BTS camera system. Across tasks, peak knee flexion angle and peak hip flexion were found to be consistent and in agreement when the Microsoft Kinect® camera system was directly compared to the BTS camera system but these values were improved following application of a corrective factor. The Microsoft Kinect® may not be an appropriate surrogate for traditional motion analysis technology, but it may have potential applications as a real-time feedback tool in pathological or high injury risk populations.  相似文献   

13.
The use of digital camcorders in biomechanical analyses can introduce errors due to inter-camera time offsets. The faster the motion being recorded the greater the error. A sequential synchronization method was developed in this study to achieve sub-field camera synchronization for multiple camcorders through numerical optimization. A recreational golfer performed ten drives while being recorded with four digital camcorders (60-Hz sampling rate). Video signals were sampled at 10,000 Hz to determine the actual inter-camera time offsets. The optimized inter-camera time offsets were computed based on three markers placed on the shaft, each separated by 200 mm. The inter-camera time offset error was computed as the difference between the optimized and actual inter-camera time offsets. The inter-camera time offset error reduced on average from 0.518 to 0.019 fields (1 field = 16.7 ms) or less due to sequential optimization. The optimized global reconstruction errors were less than 19% of the unadjusted values. It was concluded that the ability to synchronize multiple (two or more) cameras using a sequential sub-field optimization strategy promises to extend the use of relatively inexpensive digital camcorders to motions considered too fast for the low field rates of such cameras. The sequential approach presented provides a balance between computation time and reconstruction accuracy.  相似文献   

14.
Skiing‐Skating     
An articulated boot design is commonly used in skiing and skating sports because it allows sagittal plane ankle mobility while still providing critical frontal plane stability. Although articulated boots have been in use for several decades, current manufacturers of these boots differ in their articulation placement. In this study we determined an optimal position of the ankle articulation axis. We also calculated the amount of anterior skin movement that a boot tongue must account for during a full range of ankle motion. Three‐dimensional kinematic data were collected and analyzed from 40 participants moving their right foot through a full range of sagittal plane motion. The calculated horizontal position of the articulation axis was found to be highly predictable from foot length (r = 0.87, standard error of estimate = 3.44 mm), while its vertical component displayed less predictability (r = 0.49, standard error of estimate = 7.46 mm). The expansion required by the boot tongue had a moderate association with foot length and low variability (r = 0.58, standard error of estimate = 0.07 mm). An accurate axis placement will minimize relative motion between the boot cuff and the ankle, reducing friction and motion resistance. An expandable tongue will accommodate full plantar flexion and reduce pressure on the anterior ankle during dorsiflexion, eliminating common pressure‐related injuries.  相似文献   

15.

Fast unloaded movements (i.e. striking, throwing and kicking) are typically performed in a proximo‐distal sequence, where initially high proximal segments accelerate while distal segments lag behind, after which proximal segments decelerate while distal segments accelerate. The aims of this study were to examine whether proximal segment deceleration is performed actively by antagonist muscles or is a passive consequence of distal segment movement, and whether distal segment acceleration is enhanced by proximal segment deceleration. Seventeen skilled taekwon‐do practitioners were filmed using a high‐speed camera while performing a high front kick. During kicking, EMG recordings were obtained from five major lower extremity muscles. Based on the kinematic data, inverse dynamics computations were performed yielding muscle moments and motion‐dependent moments. The results indicated that thigh deceleration was caused by motion‐dependent moments arising from lower leg motion and not by active deceleration. This was supported by the EMG recordings. Lower leg acceleration was caused partly by a knee extensor muscle moment and partly by a motion‐dependent moment arising from thigh angular velocity. Thus, lower leg acceleration was not enhanced by thigh deceleration. On the contrary, thigh deceleration, although not desirable, is unavoidable because of lower leg acceleration.  相似文献   

16.
通过DLT理论的研究，结合人体运动图像三维解析系统的开发，从理论和实际应用中所出现误差原因的分析，给出实际误差修正的方法，并进一步提出续集和建议，以供实际应用参考和其它设计人员在后继研究设计中借鉴。  相似文献   

17.
Abstract

The velocities and distances covered by players during competition serve as a basis for planning fitness regimes according to the specific demand of the sport. The techniques used to calculate these movement parameters have ranged from human judgements to technological solutions such as GPS and computer vision. This paper evaluates the accuracy of a computerized motion tracking system (SAGIT/Squash) that uses computer vision methods on video captured via a fixed single camera located centrally above the court. Digital images were processed automatically with operator supervision so that any tracking errors could be rectified and manual tagging of all shots added. Four separate experiments were used to assess the error associated with tracking adult players' velocities and positions with respect to the court floor. Experiment 1 involved players standing still in different areas of the court. The tracking software was found to be more accurate when a player was stood in the centre of the court (1.33 m · min?1 error) than in the corners (2.61 m · min?1 error), predominately due to systematic errors (e.g. calibration). Experiment 2 was conducted in the same manner as Experiment 1 except that the players vigorously swung a racket around their body continuously. This resulted in 15 times the error found in Experiment 1 for the distance covered during 1 min. However, this is an unrealistic estimate of the true error when assessing matches, as during matches the racket is only swung approximately 35% of the time. Experiment 3 involved a player running at different speeds around a rectangular path on the court. The resultant trajectory, as captured by the software, was compared using different Gaussian smoothing equations of kernel widths 0.25 s, 0.5 s, and 1 s. The best solution (0.5 s) resulted in the most accurate trajectory, although the difference in distance calculated between the different equations was negligible. Experiment 4 used the 0.5-s smoothing equation to assess the tracking accuracy for a player running at a relatively steady speed in a more realistic circular trajectory. The trajectory of the pixel image was shown to have a smaller radius than the reference trajectory at increased speeds, due to the tendency of the player to lean over when negotiating a circular path. The error associated with the distance covered over 1 min was shown to range between 1.33 and 21 m depending on the nature and position of the player's movements. Values, typically somewhere in this range, are likely to be evident during typical use of this software.  相似文献   

18.
在人体运动的分析中,提取运动信息是必须的。对于人体运动的三维分析,图像系统常常是利用二维图像作三维重构,然后根据重构数据进行运动学、动力学分析。在重构技术中DLT算法由于其众多优点而得到了广泛应用。但是,由于人的运动是复杂而精细的,三维重构误差对运动分析误差的影响成为三维分析研究的关键点。针对三维重构所引起的误差,讨论了如何通过有效的方法来减小误差,以提高三维解析数据的精度,并提出了相关的方法和措施,为编制三维解析系统时提高精度提供可靠的理论依据。  相似文献   

19.
使用两部JVC摄像机对爱捷008-D辐射式框架进行定点拍摄,运用直接线性转移 (DLT)方法进行标定框架的三维重构和精度分析。结果表明:辐射式框架的控制点数目足够多且在标定空间内均匀分布时,如果控制点张成空间大于并包含被标定对象,可以达到比较好的标定精度;如果控制点均匀地分布于被标定对象的内侧和外侧,可以达到的标定精度更好。  相似文献   

20.
Propulsive forces are important determinants of swimming performance. The aim of this study was to quantify the measurement error (uncertainty) in propulsive forces calculated from kinematic data. Ten operators digitized underwater video recordings of a breaststroke swimmer's right arm action. Four landmarks on the hand were digitized at 50 Hz and their three‐dimensional coordinates obtained using a DLT algorithm. Two angles (α and ψ) defining the orientation of the hand relative to the fluid flow were calculated following the procedures of Schleihauf et al. (1983). The hydrodynamic force acting on the hand (F R) was calculated using the force coefficients of Schleihauf (1979). Errors in single measurements of hand speed, α and ψ were estimated for each video field analysed. Errors in α and ψ led to average errors in the lift and drag coefficients of 27 and 20% respectively, which, when combined with an average hand speed error of 6%, produced an average error in F R of 26%. Each of these errors was reduced by a factor of √10 when the mean of 10 measurements was used to calculate F R . Researchers should report both the estimated errors in their hydrodynamic data and the procedures used to reduce them.  相似文献   

设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号