No abstract

No abstract

Age assessment has attracted increasing attention in the field of forensics. However, most existing works are laborious and requires domain-specific knowledge. Modern computing power makes it is possible to leverage massive amounts of data to produce more reliable results. Therefore, it is logical to use automated age estimation approaches to handle large datasets. In this study, a fully automated age prediction approach was proposed by assessing 3D mandible and femur scans using deep learning. A total of 814 post-mortem computed tomography scans from 619 men and 195 women, within the age range of 20–70, were collected from the National Forensic Service in South Korea. Multiple preprocessing steps were applied for each scan to normalize the image and perform intensity correction to create 3D voxels that represent these parts accurately. The accuracy of the proposed method was evaluated by 10-fold cross-validation. The initial cross-validation results illustrated the potential of the proposed method as it achieved a mean absolute error of 5.15 years with a concordance correlation coefficient of 0.80. The proposed approach is likely to be faster and potentially more reliable, which could be used for age assessment in the future.

This research was funded by Vietnam Ministry of Science and Technology under grant number DTDLCN-16/18 ''Automated Respiration Symptoms monitoring and Abnormal Human Activity Detection Using the Internet of Things''. ABSTRACT Recently, the recent advancement of deep learning with the capacity to perform automatic highlevel feature extraction has achieved promising performance for sensor-based human activity recognition (HAR). Among different deep learning methods, Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) have been widely adopted. However, scalar outputs and pooling in CNN only allow to get the invariance but not the equivariance. The capsule networks (CapsNet) with the vector output and routing by agreement is able to capture the equivariance. In this paper, we propose a method for recognizing human activity from wearable sensors based on a capsule network named SensCapsNet. The architecture of SensCapsNet is designed to be suitable for spatial-temporal data coming from wearable sensors. Experimental results show that the proposed network outperforms CNN and LSTM methods. The performance of the proposed CapsNet architecture is assessed by altering dynamic routing between capsule layers. The proposed SensCapsNet yields improved accuracy values of 77.7% and 70.5% for 1 routing on two testing datasets in comparison with the baseline methods based on CNN and LSTM that yields the F1score of 67.7% and 69.2% for the first dataset and 65.3% and 67.6% for the second dataset respectively. Moreover, even several human activity datasets are available, privacy invasion and obtrusive concerns have not been carefully taken in to consideration in dataset building. Toward to build a non-obstructive sensing based human activity recognition method, in this paper, a dataset named 19NonSens is designed and collected from twelve subjects wearing e-Shoes and a smart watch to perform 19 activities under multiple contexts. This dataset will be made publicity available. Finally, thanks to the promising results obtained by the proposed method, we develop a life logging application which achieves a real-time computation and the accuracy rate greater than 80% for 5 common upper body activities. INDEX TERMS Human activity recognition, capsule net, wearable sensors.

No abstract