Search

Analyzing human behavior from surveillance videos can help improve public safety. Anomaly detection from time-series data is effective in facilitating this technique, and various methods based on statistical, machine learning, and deep learning techniques have been proposed. In particular, deep neural network (DNN) based methods are attracting attention because of their expressive power and ease of implementation. However, in order to apply a DNN model to time-series data analysis, it is necessary to train the DNN with a large amount of data. This not only requires time-series data obtained from an actual system but also requires data labels that describe data abnormalities. This implies a large amount of data preparation and manual annotation by humans. It is not only time consuming but also difficult to annotate data with unclear definitions, such as unnatural human behavior. In this paper, we propose a DNN model in order to automatically extract abnormal features with unclear definitions that are hidden in time-series data. We adopt a weakly supervised training method by devising a loss function to optimize the DNN model. Through experiments, we confirm that the proposed approach can be used to detect and quantify the degree of outliers in time-series data. Furthermore, unnatural human motion can be detected by applying the proposed method.

Skeleton-based human action recognition from video sequences is currently an active topic of research. Conventionally, human action recognition is performed after conducting feature extraction on a given spatial-temporal representation of a human pose by using statistical methods or deep learning methods. The spatial and temporal features are globally evaluated by a classifier and used to determine which action is closest. However, the conventional methodology does not identify the temporal location of the action that determines the classification. To address this problem, we propose a skeleton-based human action recognition and localization method using weakly supervised graph convolutional neural networks, which are both spatially and temporally connected. In this method, human action localization is accomplished using time series data of human joint positions as input and then applying regression to find an expected value for each action at each time frame. Our weakly supervised training is based on multiple-instance learning inspired by deep ranking, and we devise a loss function so that high scores can be spontaneously learned for temporally important time frames. In this paper, we first explain the network architecture and then present a multiple-instance learning method for its optimization. In the experiment, we performed localization and classification of human actions by using this method and confirmed the temporal localization efficacy of the method.

Magnetic resonance spectroscopic imaging is an emerging non-invasive technique for both anatomic and metabolic evaluation in vivo, but a primary drawback of MRSI is its long acquisition time. We propose a post-processing method to calculate MR spectra by analyzing a statistical value of images scanned at two different acquisition bandwidths. Only normal MRI pulse sequence is needed in this method and it does not require hardware specially designed for fast imaging such as EP- MRSI or SENSE. Two-dimensional analysis of water and lipid has been implemented to verify the theory. The method may provide a different approach to realize fast MRSI.

A neural network mapping approach has been proposed for the inversion problem in eddy-current testing (ECT). The use of a principal component analysis (PCA) data transformation step, a data fragmentation technique, jittering, and of a data fusion approach proved to be instrumental auxiliary tools that support the basic training algorithm in coping with the strong ill-posedness of the inversion problem. The present paper reports on the further improvements brought by a new, randomly generated database used for the training set, proposed for the reconstruction of crack shape and conductivity distribution. Good results were obtained for four levels of conductivity and nonconnected crack shapes even in the presence of high noise levels.