Martinez Julian, Decker Autumn, Cho Chi C, Doherty Aiden, Swartz Ann M, Staudenmayer John W, Strath Scott J
Department of Kinesiology, University of Wisconsin-Milwaukee.
University of Oxford and National Institute of Health Research Oxford Biomedical Research Center.
J Meas Phys Behav. 2021;4(1):47-52. doi: 10.1123/jmpb.2020-0038. Epub 2021 Feb 24.
To assess the convergent validity of body worn wearable camera (WC) still-images (IMGs) for determining posture compared with activPAL (AP) classifications.
Participants (n=16, mean age 46.7±23.8yrs, 9F) wore an Autographer WC above the xyphoid process and an AP during three, 2hr free-living visits. IMGs were captured on average 8.47 seconds apart and were annotated with output consisting of events, transitory states, unknown and gaps. Events were annotations that matched AP classifications (sit, stand and move) consisting of at least 3 IMGs, transitory states were posture annotations fewer than 3 IMGs, unknown were IMGs that could not be accurately classified, and gaps were time between annotations. For analyses, annotation and AP output were converted to one-sec epochs and matched second-by-second. Total and average length of visits and events are reported in minutes. Bias and 95% CIs for event posture times from IMGs to AP posture times were calculated to determine accuracy and precision. Confusion matrices using total AP posture times were computed to determine misclassification.
43 visits were analyzed with a total visit and event time of 5027.73 and 4237.23 minutes and average visit and event lengths being 116.92 and 98.54 minutes, respectively. Bias was not statistically significant for sitting but significant for standing and movement (0.84, -6.87 and 6.04 minutes). From confusion matrices, IMGs correctly classified sitting, standing and movement 85.69%, 54.87%, and 69.41% of total AP time, respectively.
WC IMGs provide a good estimation of overall sitting time but underestimate standing and overestimate movement time. Future work is warranted to improve posture classifications and examine IMG accuracy and precision in assessing activity type behaviors.
评估与activPAL（AP）分类相比，穿戴式摄像头（WC）静止图像（IMGs）在确定姿势方面的收敛效度。
16名参与者（平均年龄46.7±23.8岁，9名女性）在剑突上方佩戴一台Autographer WC，并在三次时长为2小时的自由生活访问期间佩戴一个AP。平均每隔8.47秒拍摄一次IMGs，并标注出由事件、过渡状态、未知和间隙组成的输出。事件是由至少3张IMGs组成的与AP分类（坐、站和移动）匹配的标注，过渡状态是少于3张IMGs的姿势标注，未知是无法准确分类的IMGs，间隙是标注之间的时间。为了进行分析，将标注和AP输出转换为每秒的时段，并逐秒匹配。访问和事件的总时长及平均时长以分钟为单位报告。计算从IMGs到AP姿势时间的事件姿势时间的偏差和95%置信区间，以确定准确性和精确性。使用总AP姿势时间计算混淆矩阵，以确定错误分类。
分析了43次访问，访问和事件的总时长分别为5027.73分钟和4237.23分钟，平均访问和事件时长分别为116.92分钟和98.54分钟。坐姿的偏差无统计学意义，但站姿和移动的偏差有统计学意义（分别为0.84、-6.87和6.04分钟）。根据混淆矩阵，IMGs分别正确分类了总AP时间的85.69%、54.87%和69.41%的坐姿、站姿和移动。
WC IMGs能很好地估计总体坐姿时间，但低估了站姿时间并高估了移动时间。未来有必要开展工作以改进姿势分类，并检验IMGs在评估活动类型行为方面的准确性和精确性。
