Feature Selection and Activity Recognition from Wearable Sensors

Pirttikangas, Susanna; Fujinami, Kaori; Nakajima, Tatsuo

doi:10.1007/11890348_39

Cited by 181 publications

(106 citation statements)

References 6 publications

Supporting

Mentioning

104

Contrasting

Unclassified

Order By: Relevance

“…Within each of these seven sets, the features were derived individually for each of the three components of the triaxial accelerometer signal. Mean and standard deviation (SD) have been used in previous studies [34] to characterize windows of accelerometer data. As an extension of this set, we defined the multiple statistics features set, which additionally included median and 25th and 75th percentile [33].…”

Section: Time-and Frequency-domain Featuresmentioning

confidence: 99%

A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

Preece

Goulermas

Kenney

et al. 2009

IEEE Trans. Biomed. Eng.

462

282

View full text Add to dashboard Cite

Abstract-Driven by the demands on healthcare resulting from the shift toward more sedentary lifestyles, considerable effort has been devoted to the monitoring and classification of human activity. In previous studies, various classification schemes and feature extraction methods have been used to identify different activities from a range of different datasets. In this paper, we present a comparison of 14 methods to extract classification features from accelerometer signals. These are based on the wavelet transform and other well-known time-and frequency-domain signal characteristics. To allow an objective comparison between the different features, we used two datasets of activities collected from 20 subjects. The first set comprised three commonly used activities, namely, level walking, stair ascent, and stair descent, and the second a total of eight activities. Furthermore, we compared the classification accuracy for each feature set across different combinations of three different accelerometer placements. The classification analysis has been performed with robust subject-based cross-validation methods using a nearest-neighbor classifier. The findings show that, although the wavelet transform approach can be used to characterize nonstationary signals, it does not perform as accurately as frequencybased features when classifying dynamic activities performed by healthy subjects. Overall, the best feature sets achieved over 95% intersubject classification accuracy.

show abstract

Section: Time-and Frequency-domain Featuresmentioning

confidence: 99%

A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

Preece

Goulermas

Kenney

et al. 2009

IEEE Trans. Biomed. Eng.

462

282

View full text Add to dashboard Cite

show abstract

“…Forward-Backward sequential search methods for feature selection has been suggested by [16], [17]. In [16] features such as mean, standard deviation, correlation (x, y axes), mean crossing, as well as heart rate mean were tested with forward-backward search, which is a well-known feature selection algorithm.…”

Section: Related Workmentioning

confidence: 99%

“…In [16] features such as mean, standard deviation, correlation (x, y axes), mean crossing, as well as heart rate mean were tested with forward-backward search, which is a well-known feature selection algorithm. With this procedure, a subset of best (giving the best classification result) fea-tures can be determined for the final analysis.…”

Section: Related Workmentioning

confidence: 99%

A Feature Set Evaluation for Activity Recognition with Body-Worn Inertial Sensors

Muhammad

Klein

Laerhoven

et al. 2012

Communications in Computer and Information Science

View full text Add to dashboard Cite

Abstract. The automatic and unobtrusive identification of user activities is a challenging goal in human behavior analysis. The physical activity that a user exhibits can be used as contextual data, which can inform applications that reside in public spaces. In this paper, we focus on wearable inertial sensors to recognize physical activities. Feature set evaluation for 5 typical activities is performed by measuring accuracy for combinations of 6 often-used features on a set of 11 well-known classifiers. To verify significance of this analysis, a t-test evaluation was performed for every combination of these feature subsets. We identify an easy-to-compute feature set, which has given us significant results and at the same time utilizes a minimum of resources.

show abstract

“…Sensors placed on both upper and lower body improve the classification accuracy [1,16]. Using other sensors than acceleration or inertial sensors was also investigated in e.g., [15], but it was shown that 3D-acceleration sensors are the most powerful sensors for activity recognition.…”

Section: Activity Recognition Using Biomechanical Model Based Pose Esmentioning

confidence: 99%

Activity Recognition Using Biomechanical Model Based Pose Estimation

Reiss

Hendeby

Bleser

et al. 2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. In this paper, a novel activity recognition method based on signal-oriented and model-based features is presented. The model-based features are calculated from shoulder and elbow joint angles and torso orientation, provided by upper-body pose estimation based on a biomechanical body model. The recognition performance of signal-oriented and model-based features is compared within this paper, and the potential of improving recognition accuracy by combining the two approaches is proved: the accuracy increased by 4-6% for certain activities when adding model-based features to the signal-oriented classifier. The presented activity recognition techniques are used for recognizing 9 everyday and fitness activities, and thus can be applied for e.g., fitness applications or 'in vivo' monitoring of patients.

show abstract

Feature Selection and Activity Recognition from Wearable Sensors

Cited by 181 publications

References 6 publications

A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

A Feature Set Evaluation for Activity Recognition with Body-Worn Inertial Sensors

Activity Recognition Using Biomechanical Model Based Pose Estimation

Contact Info

Product

Resources

About