Access to Data Sets: ERNCIP Thematic Group Video Surveillance for Security of critical infrastructure

The main objective of this report is to analyse existing data sets for video analytics (VA) and to determine how best to enable collection/common access to data sets in the EU for testing/evaluation of video surveillance software. This report presents a critical analysis of video analytic datasets with specific attention towards protection of critical infrastructures. The introductory part of the report describes the importance of video analytics and the growth of the related market. In this scenario the importance of the usage of a common dataset is highlighted. The main reason of the fundamental importance of datasets in video analysis is the intrinsic complexity of VA related techniques: a common set of video sequences is seen as a powerful boost in the design, development and test of VA algorithms. This report describes different aspects that make VA so complex and demonstrates the importance of having common and widespread dataset. Dataset must rely as well on the availability of standards related to several aspects of the video analytics for critical infrastructures protection: refer to [1] for an overview of standards in video surveillance, including the need for standards, an overview of existing relevant standardisation efforts including gaps, and a roadmap for future standards development. VA Dataset critical issues are described and analysed in details, and a simple but effective “dataset construction checklist” is proposed. In Appendix A, several existing dataset are summarized and commented in relation with the use cases highlighted in the report [2]. Moreover, the impact of each dataset in the scientific community is estimated by considering the total number of referencing papers and the most relevant research using the dataset for computing the performances of a proposed technique. With this report, we follow up on the recommendations regarding test data sets for video analytics use cases of [3] and [2]. In particular: - together with [3] and the End User Guide [4], this report helps build an argument for why datasets matter in the boardroom of critical infrastructure end users and industry; - this report gives the requirements for creating high-quality relevant data sets. Video analytics modules represent the core components of automatic video surveillance systems: these modules are able to process video sequences acquired from single or multiple video sensors, extract high-level information and automatically identify situations of interest or potentially dangerous for maintaining an appropriate level of safety for the considered environment. One of the typical key requirements from CI operators for video analytics modules is that they must guarantee a sufficient level of performance 24 hours per day, 7 days per week (24/7). Unfortunately, because of the high variability of the visual information even in a simple video surveillance installation, this simple feature is typically extremely tough to achieve. Moreover, in real video-surveillance systems an extremely wide variety of heterogeneous sensors can be found with a huge number of functionalities in ever changing scenarios. A typical approach for solving these problems is to test each video analysis module against a wide variety of video sequences: for this reason, standard datasets play a fundamental role in the design and implementation of real market-ready video surveillance systems. This report deals with video-analytics datasets, considering main features of these data and highlighting pros and cons of all the considered sets. Features are identified by considering their importance in solving the 24/7 requisites: from the results of the report it is clear that certain sequences are better suited for specific functionalities and not all the existing datasets can be used in all the real environments. The existence of some kind of ground truth for the considered dataset represents a very important feature of the dataset as it may allow an objective and quantitative evaluation of the video analytics module.JRC.E.2-Technology Innovation in Securit