Adil Mehmood Khan

A Triaxial Accelerometer-Based Physical-Activity Recognition via Augmented-Signal Features and a Hierarchical Recognizer

Physical-activity recognition via wearable sensors can provide valuable information regarding an individuals degree of functional ability and lifestyle. In this paper, we present an accelerometer sensor-based approach for human-activity recognition. Our proposed recognition method uses a hierarchical scheme. At the lower level, the state to which an activity belongs, i.e., static, transition, or dynamic, is recognized by means of statistical signal features and artificial-neural nets (ANNs). The upper level recognition uses the autoregressive (AR) modeling of the acceleration signals, thus, incorporating the derived AR-coefficients along with the signal-magnitude area and tilt angle to form an augmented-feature vector. The resulting feature vector is further processed by the linear-discriminant analysis and ANNs to recognize a particular human activity. Our proposed activity-recognition method recognizes three states and 15 activities with an average accuracy of 97.9% using only a single triaxial accelerometer attached to the subjects chest.

Human Activity Recognition via an Accelerometer-Enabled-Smartphone Using Kernel Discriminant Analysis

Nowadays many people use smartphones with built-in accelerometers which makes these smartphones capable of recognizing daily activities. However, mobile phones are carried along freely instead of a firm attachment to a body part. Since the output of any body-worn triaxial accelerometer varies for the same physical activity at different positions on a subjects body, the acceleration data thus could vary significantly for the same activity which could result in high within-class variance. Therefore, realization of activity-aware smartphones requires a recognition method that could function independent of phones position along subjects bodies. In this study, we present a method to address this problem. The proposed method is validated using five daily physical activities. Activity data is collected from five body positions using a smartphone with a built-in triaxial accelerometer. Features including autoregressive coefficients and signal magnitude area are calculated. Kernel Discriminant Analysis is then employed to extract the significant non-linear discriminating features which maximize the between-class variance and minimize the within-class variance. Final classification is performed by means of artificial neural nets. The average accuracy of about 96% illustrates the effectiveness of the proposed method.

Accelerometer signal-based human activity recognition using augmented autoregressive model coefficients and artificial neural nets

Automatic recognition of human activities is one of the important and challenging research areas in proactive and ubiquitous computing. In this work, we present some preliminary results of recognizing human activities using augmented features extracted from the activity signals measured using a single triaxial accelerometer sensor and artificial neural nets. The features include autoregressive (AR) modeling coefficients of activity signals, signal magnitude areas (SMA), and title angles (TA). We have recognized four human activities using AR coefficients (ARC) only, ARC with SMA, and ARC with SMA and TA. With the last augmented features, we have achieved the recognition rate above 99% for all four activities including lying, standing, walking, and running. With our proposed technique, real time recognition of some human activities is possible.

Accelerometer’s position independent physical activity recognition system for long-term activity monitoring in the elderly

Mobility is a good indicator of health status and thus objective mobility data could be used to assess the health status of elderly patients. Accelerometry has emerged as an effective means for long-term physical activity monitoring in the elderly. However, the output of an accelerometer varies at different positions on a subject’s body, even for the same activity, resulting in high within-class variance. Existing accelerometer-based activity recognition systems thus require firm attachment of the sensor to a subject’s body. This requirement makes them impractical for long-term activity monitoring during unsupervised free-living as it forces subjects into a fixed life pattern and impede their daily activities. Therefore, we introduce a novel single-triaxial-accelerometer-based activity recognition system that reduces the high within-class variance significantly and allows subjects to carry the sensor freely in any pocket without its firm attachment. We validated our system using seven activities: resting (lying/sitting/standing), walking, walking-upstairs, walking-downstairs, running, cycling, and vacuuming, recorded from five positions: chest pocket, front left trousers pocket, front right trousers pocket, rear trousers pocket, and inner jacket pocket. Its simplicity, ability to perform activities unimpeded, and an average recognition accuracy of 94% make our system a practical solution for continuous long-term activity monitoring in the elderly.

Accelerometer's position free human activity recognition using a hierarchical recognition model

Monitoring of physical activities is a growing field with potential applications such as lifecare and healthcare. Accelerometry shows promise in providing an inexpensive but effective means of long-term activity monitoring of elderly patients. However, even for the same physical activity the output of any body-worn Triaxial Accelerometer (TA) varies at different positions of a subjects body, resulting in a high within-class variance. Thus almost all existing TA-based human activity recognition systems require firm attachment of TA to a specific body part, making them impractical for long-term activity monitoring during unsupervised free living. Therefore, we present a novel hierarchical recognition model that can recognize human activities independent of TAs position along a human body. The proposed model minimizes the high within-class variance significantly and allows subjects to carry TA freely in any pocket without attaching it firmly to a body-part. We validated our model using six daily physical activities: resting (sit/stand), walking, walk-upstairs, walk-downstairs, running, and cycling. Activity data is collected from four most probable body positions of TA: chest pocket, front trousers pocket, rear trousers pocket, and inner jacket pocket. The average accuracy of about 95% illustrates the effectiveness of the proposed method.

A single tri-axial accelerometer-based real-time personal life log system capable of activity classification and exercise information generation

Recording a personal life log (PLL) of daily activities is an emerging technology for u-lifecare and e-health services. In this paper, we present an accelerometer-based personal life log system capable of human activity classification and exercise information generation. In our system, we use a tri-axial accelerometer and a real-time activity recognition scheme in which a set of augmented features of accelerometer signals, processed with Linear Discriminant Analysis (LDA), is classified by our hierarchical artificial neural network classifier: in the lower level of the classifier, a state of an activity is recognized based on the statistical and spectral features; in the upper level, an activity is recognized with a set of augmented features including autoregressive (AR) coefficients, signal magnitude area (SMA), and tilt angles (TA). Upon the recognition of each activity, we further estimate exercise information such as energy expenditure based on Metabolic Equivalents (METS), step count, walking distance, walking speed, activity duration, etc. Our PLL system functions in real-time and all information generated from our system is archived in a daily-log database. By testing our system on seven different daily activities, we have obtained an average accuracy of 84.8% in activity recognition and generated their relative exercise information.

Extraction of P300 using constrained independent component analysis

A brain computer interface (BCI) uses electrophysiological activities of the brain such as natural rhythms and evoked potentials to communicate with some external devices. P300 is a positive evoked potential (EP), elicited approximately 300ms after an attended external stimulus. A P300-based BCI uses this evoked potential as a means of communication with the external devices. Until now this P300-based BCI has been rather slow, as it is difficult to detect a P300 response without averaging over a number of trials. Previously, independent component analysis (ICA) has been used in the extraction of P300. However, the drawback of ICA is that it extracts not only P300 but also non-P300 related components requiring a proper selection of P300 ICs by the system. In this study we propose an algorithm based on constrained independent component analysis (cICA) for P300 extraction which can extract only the relevant component by incorporating a priori information. A reference signal is generated as this a priori information of P300 and cICA is applied to extract the P300 related component. Then the extracted P300 IC is segmented, averaged, and classified into target and non-target events by means of a linear classifier. The method is fast, reliable, computationally inexpensive as compared to ICA and achieves an accuracy of 98.3% in the detection of P300.

MUQAMI: A Locally Distributed Key Management Scheme for Clustered Sensor Networks

In many of the sensor network applications like natural habitat monitoring and international border monitoring, sensor networks are deployed in areas, where there is a high possibility of node capture and network level attacks. Specifically in such applications, the sensor nodes are severely limited in resources. We propose MUQAMI, a locally distributed key management scheme for resilience against the node capture in wireless sensor networks. Our scheme is efficient both in case of keying, re-keying and node compromise. Beauty of our scheme is that it requires minimal message transmission outside the cluster. We base our Scheme on Exclusion Basis System (EBS).

Comparative Analysis of XLMiner and Weka for Association Rule Mining and Clustering

Retaining a customer is preferred more than attracting new customers. Business organizations are adopting different strategies to facilitate their customers in verity of ways, so that these customers keep on buying from them. Association Rule Mining (ARM) is one of the strategies that find out correspondence/association among the items sold together by applying basket analysis. The clustering technique is also used for different advantages like; recognizing class of most sold products, classifying customers based on their buying behavior and their power of purchase. Different researchers have provided different algorithms for both ARM and Clustering, and are implemented in different data mining tools. In this paper, we have compared the results of these algorithms against their implementation in Weka and XLMiner. For this comparison we have used the transaction data of Sales Day (a super store). The results are very encouraging and also produced valuable information for sales and business improvements.

An efficient re-keying scheme for cluster based wireless sensor networks

Due to vast application of WSN (Wireless Sensor Networks) in mission critical military operations, securing WSN has received lot of attention from the research community. WSN when deployed in hostile environment, they are prone to various kinds of attacks one of which is node capture which might reveal important sensor information being transferred to the captured node. Thus dynamic key management schemes employ re-keying mechanism to change the group key used by the sensor nodes for communication. Constrained resources such as energy, memory and computational capabilities of sensor nodes requires a re-keying scheme efficient in design to minimize overhead while maintaining secure communications over lifespan of the network. In this paper we present an efficient re-keying scheme for cluster based WSN which requires minimal communication with the base station and O(1) computation at the sensor node to calculate the new group key.