Sistema Context-Aware de Videovigilancia Inteligente bajo el paradigma Edge-Computing

Abstract

Nowadays, video surveillance systems are one of the main tools for the prevention, detection and investigation of crimes against public security. As a result, the global video surveillance market has grown steadily in recent decades. According to the latest reports, this growth is expected to continue over the next few years, with an expected annual growth of almost 17 %. Traditional video surveillance systems consist of several cameras connected to monitors and/or recorders. For these systems to add value, it is vitally important that there are security personnel watching the monitors or reviewing the recordings. However, on the one hand, there are more and more cameras, which makes monitoring difficult. On the other hand, it has been shown that a person quickly loses attention when performing this type of activity. The latest advances in video surveillance try to mitigate these problems by automating a large part of the activities carried out by security professionals. The field of study that studies these automations is called intelligent video surveillance and was born just over two decades ago. Today, both a traditional video surveillance system and an intelligent video surveillance system use a wide variety of technologies for recording, sending, managing and displaying video signals. However, intelligent video surveillance systems need extra technology for processing video signals. These technologies are related to artificial intelligence and, above all, artificial vision. Intelligent video surveillance has grown a lot in recent years due to several factors. On the one hand, there is a growing demand from security personnel, who cannot cover all needs. On the other hand, computers have grown enormously in computing capabilities, which in turn has allowed to improve the artificial vision algorithms needed in this field. All together has led to an ecosystem both at the research level and at the commercialization level that has allowed a great advance in a short period of time in the matter. The latest advances in intelligent video surveillance seem to be related to adding as much information as possible to these systems. The main objective is for them to be able to make better decisions and for these decisions to be more complex. In this line, the use of context information is framed, giving rise to what is known as Context-Aware systems. The use of context information has already proved to be very useful in this context, as these systems have more intelligence and flexibility than traditional systems. That is why it has started to be applied in different intelligent video surveillance systems. Although the use of context information can be very useful, adding even more information to systems that already create very large volumes of data can create problems in the management of these. There are several solutions and paradigms at the level of architecture and distribution of processes to solve these problems. In this sense, one of the paradigms that is taking more presence in this sector is the Edge-Computing. This paradigm of computation distribution says that the processing must be as close as possible to the place where the data originates. It is a paradigm closely related to the Internet of Thing, a concept that in turn has a great relationship with intelligent video surveillance. Both Context-Aware and Edge-Computing have already been used in the field of intelligent video surveillance. However, no application or system has been found in the scientific literature where they occur at the same time. However, given the results they have already demonstrated separately, it seems logical to think that combining them could create even more advanced systems. A system with these characteristics would obtain great scalability due to the Edge-Computing paradigm, and great flexibility and decision-making capacity due to Context-Aware. This is the main motivation of this thesis. It is intended to create an intelligent video surveillance system that combines both concepts. In order to do this, we first present an architecture based on the Edge-Computing paradigm that also supports context information. Then two different algorithms are proposed that can be integrated into this architecture and that make use of this context information. Finally, a context-conscious risk methodology is proposed to manage and summarize all the alarms generated by the system. This method allows to add all the useful information for the security personnel in an easily understandable risk signal.