Anomaly Detection in Videos Using Two-Stream Autoencoder with Post Hoc Interpretability

The growing interest in deep learning approaches to video surveillance raises concerns about the accuracy and efficiency of neural networks. However, fast and reliable detection of abnormal events is still a challenging work. Here, we introduce a two-stream approach that offers an autoencoder-based structure for fast and efficient detection to facilitate anomaly detection from surveillance video without labeled abnormal events. Furthermore, we present post hoc interpretability of feature map visualization to show the process of feature learning, revealing uncertain and ambiguous decision boundaries in the video sequence. Experimental results on Avenue, UCSD Ped2, and Subway datasets show that our method can detect abnormal events well and explain the internal logic of the model at the object level.

Smart-Object-Based Reasoning System for Indoor Acoustic Profiling of Elderly Inhabitants

Many countries are facing significant challenges in relation to providing adequate care for their elderly citizens. The roots of these issues are manifold, but include changing demographics, changing behaviours, and a shortage of resources. As has been witnessed in the health sector and many others in society, technology has much to offer in terms of supporting people’s needs. This paper explores the potential for ambient intelligence to address this challenge by creating a system that is able to passively monitor the home environment, detecting abnormal situations which may indicate that the inhabitant needs help. There are many ways that this might be achieved, but in this paper, we will describe our investigation into an approach involving unobtrusively ’listening’ to sound patterns within the home, which classifies these as either normal daily activities, or abnormal situations. The experimental system we built was composed of an innovative combination of acoustic sensing, artificial intelligence (AI), and the Internet-of-Things (IoT), which we argue in the paper that it provides a cost-effective approach to alerting care providers when an elderly person in their charge needs help. The majority of the innovation in our work concerns the AI in which we employ Machine Learning to classify the sound profiles, analyse the data for abnormal events, and to make decisions for raising alerts with carers. A Neural Network classifier was used to train and identify the sound profiles associated with normal daily routines within a given person’s home, signalling departures from the daily routines that were then used as templates to measure deviations from normality, which were used to make weighted decisions regarding calling for assistance. A practical experimental system was then designed and deployed to evaluate the methods advocated by this research. The methodology involved gathering pre-design and post-design data from both a professionally run residential home and a domestic home. The pre-design data gathered the views on the system design from 11 members of the residential home, using survey questionnaires and focus groups. These data were used to inform the design of the experimental system, which was then deployed in a domestic home setting to gather post-design experimental data. The experimental results revealed that the system was able to detect 84% of abnormal events, and advocated several refinements which would improve the performance of the system. Thus, the research concludes that the system represents an important advancement to the state-of-the-art and, when taken together with the refinements, represents a line of research which has the potential to deliver significant improvements to care provision for the elderly.