Raquel Dosil | Researcher
A recent survey performed by IBM reveals alarming conclusions about the traffic in big cities, like for example, that one third of the total traffic corresponds to vehicles trying to find a parking place. This is just one example of the serious difficulties that big cities are facing in the management of their services and resources. These issues are pushing local administrations towards the adoption of measures that allow the improvement of the habitability and efficiency of the cities, giving place to the concept of Smart City.
One of the strategies to achieve more efficient cities consists in capturing large amounts of detailed data about what is happening in the city at each time instant, and then analyzing it to extract useful high level information to support decision making and process control tasks. Following the example of the urban mobility, if detailed information about the traffic of a city is available, relevant conclusions may be taken about traffic flow direction selection, signposting, planning of new routes, etc.
In this sense, real time video analytics shows up as a suitable source of information. In the one hand, video provides information that is directly understandable by a human operator, what allows easy monitoring of large areas. On the other side, temporal sequences of images contain a large amount of information compared to other types of sensors, and this information can be automatically extracted by means of video analytics software. Thus, by incorporating the appropriate algorithms, the same camera can act as many diverse sensors, such as sensors for detection of motion, lightness, objects –like cars and people–, and events –like traffic violations or accidents– etc.
Nevertheless, the introduction of video analytics in Smart City applications has to face important challenges:
• Scalability: high demand of network, processing, storage and energy resources.
• Reliability: sensitivity to changes in capturing conditions and to intrinsic variability of the observed real world.
• Ease of deployment: requires configuration of devices, streaming protocols, and particularly, processing algorithms.
• Speed: high computational cost
• Privacy: it must comply with privacy preserving laws.
Despite these challenges, Gradiant believes in video analytics as a versatile solution for many applications within the Smart City. For this reason, we work on several research lines devoted to endow video analytics with the required features of scalability, reliability, usability, speed, and confidentiality. These research lines can be grouped in two big goals.
The first one is focused on the optimization of the use of network resources, and includes the use of compression standards, the optimization of the camera network dimensioning, and the adaptive configuration and prioritization of video streams according to network context data. The second group is centered in the development of video analysis algorithms that are efficient, robust, self configurable and privacy preserving, and includes techniques for parallelization and decentralization of processing, self calibration, encryption of image regions with sensitive information, and development of robust detection algorithms that cope with data variability.
Gradiant has already developed a prototype of a real time video analytics system for IP camera networks. This prototype has been recently integrated in a pilot smart city management platform, deployed in a realistic scenario in the city of Barcelona, and it is currently under testing.