high performance computing on graphics processing units: hgpu.org

This thesis brings together productivity and risk assessments through innovative design, development and evaluation of a unique system for retrieving and analysing data. In the past, although the link between them is well-documented, these assessments have largely been dealt with as separate antagonist entities. A broad evaluation of the existing traditional and technological support systems has been conducted to identify suitable methodologies along with a common technological platform for automation. The methodologies selected for the productivity and risk assessments were; work sampling and the revised NIOSH lifting equation respectively. The automation of these procedures is facilitated through computer vision and the use of a range imaging Kinect™ camera. The standalone C++ application integrates two tracking approaches to extract real-time positional data on the worker and the work-piece. The OpenNI and OpenCV libraries are used to perform skeletal tracking and image recognition respectively. The skeletal tracker returns positional data on specific joints of the worker, while the image recognition component, a SURF implementation, is used to identify and track a specific work-piece within the capture frame. These tracking techniques are computationally expensive. In order to enable real time execution of the program, Nvidia’s CUDA toolkit and threading building blocks have been applied to reduce the processing time. The performance measurement system is a continuous sampling derivative of work sampling. The speed of the worker’s hand movements and proximity to the work-piece are used to classify the worker in one of four possible states; busy, static, idle, or out of frame. In addition to the worker based performance measures, data relating to work-pieces are also calculated. These include the number of work-pieces processed by a specific worker, along with the average and variations in the processing times. The risk assessment is an automated approach of the revised NIOSH lifting equation. The system calculates when a worker makes and/or breaks contact with the work-piece and uses the joint locations from the skeletal tracker to calculate the variables used in the determination of the multipliers and ultimately the recommended weight limit and lifting index. The final calculation indicates whether the worker is at risk of developing a musculoskeletal disorder. Additionally the information provided on each of the multipliers highlights which elements of the lifting task contribute the most to the risk. The user-interface design ensures that the system is easy to use. The interface also displays the results of the study enabling analysts to assess worker performance at any time in real time. The automated system therefore enables analysts to respond rapidly to rectify problems. The system also reduces the complexity of performing studies and it eliminates human errors. The time and costs required to perform the studies are reduced and the system can become a permanent fixture on factory floors. The development of the automated system opens the door for further development of the system to ultimately enable more detailed assessments of productivity and risk.