On the Correlation of Sensor Location and Human Activity Recognition in Body Area Networks (BANs)

Convolutional neural networks as means to identify apposite sensor combination for human activity recognition

“…x The performance of the proposed CNN based approach in terms of recognition accuracy was observed significantly better than several classifiers, such as the SLR, Naïve Bayes, and the SMO. On an average, the CNN has shown 26% higher Fmeasure score for recognition of thirteen different human activities than different classifier used with LESH based approach in [4].…”

“…Therefore, only an effective and complex feature extraction technique can be applied to such applications that have the capability to capture any salience feature in the dataset. In our previous work in [4], we presented a cloud based framework to investigate the effects of sensor location on the accuracy of human activity recognition. The aforementioned framework utilizes a Local Energy-based Shape Histogram (LESH) approach for transformation of human activities data into a feature space that subsequently is used by machine learning algorithms for classification of activities.…”

“…In the previous LESH based approach [4], there was not a single combination of sensors that could identify all of the activities with high accuracy. Therefore, an important question was whether the CNN based model has the capability to identify all of the activities with a single combination of sensors.…”

“…To evaluate the performance of the CNN based approach for human activity recognition, we conducted experiments on Wearable Action Recognition Database (WARD) [5] dataset comprising of the data from 20 subjects for thirteen activities using five sensors on the human body. For the details of the location of sensors on human body, interested readers are encouraged to refer to [4]. With the presented CNN based approach, the recognition accuracy for all of the activities in general and particularly for the three activities, such as walk forward, walk forward left circle, and walk forward right circle that in fact was significantly low using the LESH based approach is significantly improved.…”

An Efficient Activity Detection System based on Skeleton Joints Identification

“…The work of Jain and Kanhangad [17] has used gyroscope and accelerometers to address the classification problems in human activity recognition system. Correlation-based approach was found in Khan et al [18] to minimize the dimensional attributes involved in it along with usage of feature vector. Manzi et al [19] have offered a Skeleton-based human activity identification system.…”