This paper addresses the automated identification of violent acts from CCTV video streams using a Deep Learning model under constrained resources. While this process typically involves a powerful setup, it is useful to accelerate the training and get accurate results using more modest computational resources that would bring automatic recognition of violent acts closer to common surveillance resources. Our results provide 94.98% accuracy, on par with the state-of-the-art, but at a fraction of the training time. This translates into lower energy requirements and allows a broader deployment on large scale (urban) autonomous surveillance networks while providing an increased privacy towards citizens and lower chances of abuse from authorities.

Gesture identification represents one way of monitoring adherence to medical treatment for cognitive-impaired individuals, dementia-related conditions being severely dependent on precise medication. This paper proposes a gesture identification algorithm used to detect the pill ingestion, that runs on an inexpensive smart wearable device. We use techniques pertaining to supervised machine learning and the data set is processed with the Keras framework. Data collected is represented by acceleration values supplied by the wearable and is proceed on the wearable device itself, showing high accuracy results in identifying the pill intake gesture. The trained model is deployed in a resource constrained embedded device and the inferences is carried locally onto the device.

By using data gathered from real-life traffic situations combined with concepts of machine learning, the purpose of our work is to determine patterns and, consequently, try to predict future traffic congestion caused by anomalies such as accidents. The results will be compared to the actual data in order to assess the accuracy of the implemented solution.

Configuring the traffic light scheduling for a complex intersection in dense urban environments can be challenging because of conflicting directions that have to receive fair servicing. While commercial traffic simulations allow these configurations be computed automatically, there is often a sense of missing optimization that could be implemented and which may yield additional traffic flow. This paper presents a case study for a major intersection in the city of Timisoara, in which traffic light coordination is achieved by implementing the Firefly algorithms, its performance being compared through simulations in VISSIM with the actual conditions for the same intersection. Results show that the intersection could still benefit from additional improvements to the traffic flow.

This dataset comprises street-level traces of traffic flow as reported by Here Maps™ for 13 cities of Romania from 15th. of May 2020 and until 5th. of June 2020. This covers the time two days before lifting of the mobility restrictions imposed by the COVID19 nation-wide State of Emergency and until four days after the second wave of relaxation, announced for 1st. of June 2020. Data were sampled at a 15-min interval, consistent with the Here API update time. The data are annotated with relevant political decisions and religious events which might influence the traffic flow. Considering the relative scarcity of real-life traffic data, one can use this data set for micro-simulation during development and validation of Intelligent Transportation Solutions (ITS) algorithms while another facet would be in the area of social and political sciences when discussing the effectiveness and impact of statewide restriction during the COVID19 pandemic.

Our investigation is geared towards using traffic micro-simulation tools (PTV Vissim) fol developing and validating a genetic algorithm used to compute the green cycles in urban road environments. Classic approaches relate the theoretical models and computations but are lacking the necessary degree of fidelity with the real world because they ignore the ripple effects caused by slight disturbances in traffic and the particularities of the road configuration itself. Using micro-simulation we are capable of measuring the fitness parameters in a virtualized but highly accurate environment (detailed road network of Timisoara) and the results are going to used further for optimizing the existing traffic signaling system trough the partnership with the local authorities.

Designing and deploying intelligent transportation systems in a city always requires finding the important intersections, which is itself a difficult and subjective task. Using real-time data collected from a multitude of data sources we propose a new data fusion method and introduce a new metric for evaluating node importance based on the traffic volume. We correlate this metric with the one of network betweenness, proving the possibility of using the latter with good enough results for practical applications.

We use the network betweeness metric to rank the nodes of a network and find hot spots, prone to congestion in a road network. By applying complex networks methodologies and metrics we identify a correlation between the two and validate it by using modern road traffic simulation tools. We also provide an empirical assessment based on the authors' insight while using the road infrastructure of the targeted city.

At present, most of the attempts to quantify urban traffic flow have completely ignored its predominant social footprint, while literature concerning the appearance and forecasting of congestion points remains scarce. This paper presents a methodology for determining the critical point leading to congestion in urban traffic, based on analyzing the network topology and flow using an innovative mixture of complex network metrics and state of the art traffic simulation. To validate the concepts and methodology we conducted a case study over the city of Timisoara. At around 0.5 million inhabitants we believe it provides, which provides a suitable benchmark for deriving empirical, results and qualitative interpretations.

We present a complete technical solution for continuously monitoring vital signs required for observing sleep apnoea events, one of the major sleep respiratory disorders. Based on industry accepted medical devices, we developed a GSM-based remote data acquisition and transfer module that is integrated via a set of web services into the server side of the application. The back-end is responsible with aggregating all the data, and, based on machine learning techniques, it provides a first level of filtering in order to warn about possible abnormalities. The proposed solution is currently under the test phase at the "Victor Babes" Hospital in Timisoara, Romania.

A rapidly rising number of civilian and military real-world applications require deployments of large sensor networks. However, problems like limited energy supply, tough environments, data latency, and integrity cause adverse effects on large topologies of sensors. This paper presents a novel approach in designing the placement of relay nodes in a sensor network. By using concepts from the area of social network analysis and mapping them to the already classical field of sensor networks we succeed to add improvements to the costs implied with deploying the infrastructure. By socializing the topology with the concepts of centrality and community structure, our research is focused around a flexible design space exploration algorithm that we have devised, which offers a balance between the performance and cost of deploying relays in a sensor network. As a result, our WSN design achieves a relevant improvement over the state of the art solutions.

Motivated by the constantly growing interest and real-world applicability shown in complex networks, we model and optimize the network formed by road networks in cities from an innovative perspective. We detect traffic hotspots which lead to congestion using the betweenness centrality of the road graph. This is shown to have a power-law distribution which we set out to redistribute and equalize. Optimization at a macro-level is not feasible because of the graph size, and thus we recursively narrow down the methodology to a sub-optimization of city neighborhoods. To that end, the paper introduces a genetic algorithm which redistributes betweenness optimization at a neighborhood level, district level, and city level to reduce and/or eliminate congestion hotspots, by changing street directions, without adding any new roads. Experimental results yield an improvement with a factor of 4 times in terms of reducing load off from hotspots and transferring it to neighboring streets.

In present days, the road network in any major city faces the constant pressure of accommodating an ever increasing number of vehicles while conserving a congestion-free status. However, identifying key intersections that will soon become congested is a difficult task, performed by tedious, thorough simulations; even more difficult is to adapt the road network so as to increase its efficiency and avoid congestion. We argue upon the social component of road traffic and propose an alternative way to detect hotspots leading to congestion by using techniques borrowed from complex network analysis. We will use the betweenness centrality and argue upon its power-law distribution, which we set out to redistribute and equalize. The paper introduces a genetic algorithm that redistributes the betweenness values at a community level in a city by changing street directions and number of available lanes in order to reduce (and possibly even eliminate) congestion hotspots. Experimental results in terms of reducing traffic loads from hotspots and transferring to neighboring streets yield an improvement with a factor of up to 90% times without adding significant costs or modifying the existing infrastructure.

We present a novel approach in designing and deploying traffic light systems by identifying key intersections of the road network. Based on techniques borrowed from Complex Network Analysis, our algorithm can be applied successively at different levels of granularity allowing a hierarchical clustering of the intersections and prioritization of the traffic lights. We illustrate our approach with a case study conducted over the city of Timisoara.

We present a collaborative, distributed solution for gathering high quality quasi-real-time data regarding the conditions on public roads in an urban community by using a mobile sensor network implemented on smart-phones. Our solution for collecting data is based on a series of mobile nodes running a custom application and a Java server responsible for aggregating and processing all the data and building a state of the roads at a specific moment. Users are presented with a web interface for querying the data in a geo-enabled manner. A specialized algorithm is developed an presented for collecting data in absence of GIS information.

This paper describes an adaptive algorithm that can be used to optimize traffic movements by controlling the traffic signal operations in an intersection and sets the framework for adjacent intersections. Inter-traffic signal communication is used to respond to traffic changes by deriving new timings. We illustrate the proposed solution through a case study conducted over the city of Timisoara, Romania. Our algorithm was tested using the VISSIM simulator and results show improvements in reducing waiting times and queue lengths over the currently deployed solution based on fixed time plans.

This paper presents a modif Selective Fault Tolerance method, substantial area reduction over the stat technique is proved to achieve substant and better reliability results when appl with the so-called SAM method for relia The simulation results show that improvement of up to over 20% in terms overhead, compared with the state of the to a classic TMR we obtain improveme with a mean improvement of 25% in t energy reduction. Furthermore when simplified selective fault tolerance techn method for cache memories and compa classic TMR, not only we obtain a 65% d energy overhead, but also we manag improvement in reliability.

This paper presents a modified version of the Selective Fault Tolerance method, which achieves substantial area reduction over the state of the art. This technique is proved to achieve substantial area reduction and better reliability results when applied in conjunction with the so-called SAM method for reliable cache memory. The simulation results show that we achieved an improvement of up to over 20% in terms of area and energy overhead, compared with the state of the art. Also compared to a classic TMR we obtain improvements of up to 65%, with a mean improvement of 25% in terms of area and energy reduction. Furthermore when we combine the simplified selective fault tolerance technique to the SAM method for cache memories and compare the results to a classic TMR, not only we obtain a 65% decrease in area and energy overhead, but also we manage to achieve an improvement in reliability.