An Integrated Fusion Engine for Early Threat Detection Demonstrated in Public-Space Trials.

Counter terrorism is a huge challenge for public spaces. Therefore, it is essential to support early detection of threats, such as weapons or explosives. An integrated fusion engine was developed for the management of a plurality of sensors to detect threats without disrupting the flow of commuters. The system improves security of soft targets (such as airports, undergrounds and railway stations) by providing security operators with real-time information of the threat combined with image and position data of each person passing the monitored area. This paper describes the results of the fusion engine in a public-space trial in a metro station in Rome. The system consists of 2D-video tracking, person re-identification, 3D-video tracking, and command and control (C&C) formulating two co-existing data pipelines: one for visualization on smart glasses and another for hand-over to another sensor. Over multiple days, 586 commuters participated in the trial. The results of the trial show overall accuracy scores of 97.4% and 97.6% for the visualization and hand-over pipelines, respectively, and each component reached high accuracy values (2D Video = 98.0%, Re-identification = 100.0%, 3D Video = 99.7% and C&C = 99.5%).

Lightweight Adaptation of Classifiers to Users and Contexts: Trends of the Emerging Domain.

Intelligent computer applications need to adapt their behaviour to contexts and users, but conventional classifier adaptation methods require long data collection and/or training times. Therefore classifier adaptation is often performed as follows: at design time application developers define typical usage contexts and provide reasoning models for each of these contexts, and then at runtime an appropriate model is selected from available ones. Typically, definition of usage contexts and reasoning models heavily relies on domain knowledge. However, in practice many applications are used in so diverse situations that no developer can predict them all and collect for each situation adequate training and test databases. Such applications have to adapt to a new user or unknown context at runtime just from interaction with the user, preferably in fairly lightweight ways, that is, requiring limited user effort to collect training data and limited time of performing the adaptation. This paper analyses adaptation trends in several emerging domains and outlines promising ideas, proposed for making multimodal classifiers user-specific and context-specific without significant user efforts, detailed domain knowledge, and/or complete retraining of the classifiers. Based on this analysis, this paper identifies important application characteristics and presents guidelines to consider these characteristics in adaptation design.

Activity classification using realistic data from wearable sensors.

Automatic classification of everyday activities can be used for promotion of health-enhancing physical activities and a healthier lifestyle. In this paper, methods used for classification of everyday activities like walking, running, and cycling are described. The aim of the study was to find out how to recognize activities, which sensors are useful and what kind of signal processing and classification is required. A large and realistic data library of sensor data was collected. Sixteen test persons took part in the data collection, resulting in approximately 31 h of annotated, 35-channel data recorded in an everyday environment. The test persons carried a set of wearable sensors while performing several activities during the 2-h measurement session. Classification results of three classifiers are shown: custom decision tree, automatically generated decision tree, and artificial neural network. The classification accuracies using leave-one-subject-out cross validation range from 58 to 97% for custom decision tree classifier, from 56 to 97% for automatically generated decision tree, and from 22 to 96% for artificial neural network. Total classification accuracy is 82 % for custom decision tree classifier, 86% for automatically generated decision tree, and 82% for artificial neural network.