Human body motion can be captured by body area sensor networks. Accurate sensor placement with respect to anatomical landmarks is one of the main factors determining the accuracy of motion-capture systems. Changes in position of the sensors cause increased variability in the motion data, so isolating the characteristic features that represent the most important motion patterns is our concern. As accurate sensor placement is time-consuming and hard to achieve, we propose a signal processing technique that can enable salient data to be isolated. By using functional principal component analysis (f-PCA), we compensate for the variation in data due to changes in the on-body positioning of sensors. More precisely, we investigate the use of f-PCA for filtering and interpreting motion data, whilst accounting for variability in the sensor origin. Data are collected through a marker-based motion capture system from two designed experiments based on human body and robot arm movement. Results show differences between similar actions across different sessions of marker wearing with random changes in position of sensors. After applying the f-PCA filter on the data, we show how uncertainties due to sensor position changes can be compensated for.

No abstract

No abstract

Motion capture coupled with on-body sensing and biofeedback are key enabling technologies for assisted motor rehabilitation.However, wearability, power efficiency and measurement repeatability remain the principle challenges that need to be addressed before widespread adoption of such systems becomes possible. The weight and the size of the on-body sensing system needs to be kept small, and the system should not interfere with the user's movements or actions, but in general they are bulky due to their power consumption requirements. Furthermore, on-body sensors are very sensitive to positioning, which causes increased variability in the motion data. Isolating the characteristic patterns that represent the most important motion data affected by random positioning errors, while also reducing the power consumption, is our main concern. We consider an automated computational approach to address the two problems. We investigate the use of f-PCA for signal separation, whilst accounting for variability in the sensor position. In the designed experiments, we use human subjects and a robot arm to generate motion data, which is analogous to the human joint flexion-extension motion. The data are captured by an active marker-based motion capture system. As both the motion capture system and the robot arm are very accurate in their operation, we are able to introduce deliberate placement errors in a precisely controlled manner. The results are independent from the technology used to measure motion because we consider joint angles as variables in our analysis. The proposed approach can thus be applied to other motion capture systems. The proposed post-processing technique can compensate for uncertainties due to sensor positional changes, whilst allowing greater energy efficiency of the sensors, thus enabling improved flexibility and usability of on-body sensing.

Abstract. Performance measurement of packet (e.g. IP) networks is a vital element in the commercial viability of broadband. Active measurement by injection of probing packets will be very widely used as a method of performance measurement. In this paper we quantify the error when using probing. We discover that, when measuring the mean packet delay across a WAN through a representatively loaded (i.e. 50%-90% utilised) access link, the measurements often have an error of many tens, hundreds or even thousands of milliseconds. Furthermore, we apply results from queueing analysis to show that probing for packet loss will require that the probes be about the same size as the data packets; if small packets are used the measured packet loss probability will be many orders of magnitude smaller than it actually is. When this is accounted for, i.e. by using probing packets the same size as the data packets, then, for constant probing load (i.e. a smaller number of larger packets), the error in the returned delay measurements becomes considerably worse.