FEATURE Dynamic Testing of Civil Engineering Structures Series PEOPLE-INDUCED VIBRATIONS OF CIVIL STRUCTURES: IMAGE-BASED MEASUREMENT OF CROWD MOTION by A. Caprioli, S. Manzoni, and E. Zappa S tructural health monitoring of large civil structures hosting a lot of people, such as stadia, is getting a high consideration from the international scientific community, especially considering safety and structure safeguard. The presence of huge crowds significantly influences the structural dynamics for different reasons. The crowd motion may cause a significant excitation of the structure, 1–3 which can become almost harmonic in case of a collective movement synchronized by music or choruses, as experimentally observed in various concerts in different countries 4 and in laboratory tests with a lot of people. 5 Then the crowd constitutes a dynamic element, interacting with the structure and thus modifying its behavior, as found by Reynolds et al. 6 in the analysis of the dynamic behavior of the Bradford stadium. The two mentioned effects may lead to dangerous vibration conditions, potentially resulting in crowd panic or structural damages. A well-known example is the Bruce Springsteen concert at the Ullevi Stadium in 1985, where the excitation due to crowd motion, synchronized by the music beat, caused irreparable structural damages. A good improvement to prevent risky situations is an automated early detection system, allowing the prompt identification of possible dangerous conditions on the grandstands and to take the most appropriate safety action. At the moment, the only available strategy is related to vibration measurement. 7 This approach, however, is only related to the structural vibration monitoring, without considering crowd movement, which is the cause of the oscillation. This is because in the past no techniques were available to estimate crowd behavior. This information could be very precious, because many authors developed numerical models to estimate the people-induced loading; all these models need the crowd motion as input but usually only empirical information is available. Loads due to people motion are not directly measurable; there is thus a need to evaluate crowd movement (mainly the vertical motion) to have an excitation estimation through numerical models, already present in literature. Two examples of these models can be found in Refs. 8 and 9. Other models, more concerned with crowd motion in stadia, have been developed by Sim et al. 10 and by Dougill. 11 A comprehensive review can be finally found in Ref. 12. As Editor’s Note: This article is part of the Feature Series on Dynamic Testing of Civil Engineering Structures. This series will cover a wide range of technologies appropriate to civil engineering structures from both practical/technical and analytical perspectives. Series editor: Paul Reynolds, The University of Sheffield. Dr. A. Caprioli is a lecturer and Dr. S. Manzoni (stefano.manzoni@polimi.it) and Dr. E. Zappa are assistant professors in the Department of Mechanics, Politecnico di Milano, Via La Masa, Milan, Italy shown in the mentioned papers, the dynamic excitation of civil structures due to jumping people is mainly due to the vertical motion (jumping or bobbing) of the spectators, whereas horizontal motion (walking) produces a negligible dynamic load. Therefore, it is of practical interest to design a measurement technique capable of estimating the vertical motion of the spectators, thus allowing the prediction of loads generated by the people on the hosting structures. One of the most promising strategies is the image processing of movies depicting people on the grandstands and acquired during concerts or sport events. The problem of people tracking in image sequences is well known in literature. 13,14 Almost all the image processing approaches, proposed in the literature to measure the motion of jumping people, require a reasonably good image resolution and show some draw- backs in the case of very crowded areas; unfortunately, the spectators are usually very crowded on the stadia stands and the image resolution is frequently very poor (this is due to the need of imaging as many people as possible with a single camera). Accordingly, the use of the mentioned techniques to analyze the images of people on the stands is probably not the best solution. Moreover, because the most dangerous condi- tions are met when almost all the people jump synchronously, the individual tracking of each single person is not really needed. The motion of each person is very similar to the motion of his neighbors so that a measurement of the average motion of small groups of people gives almost the same infor- mation as provided by the measurement of the motion of each single person. Thanks to this favorable scenario, it is pro- posed to use an extremely robust image processing technique, based on area correlation matching. The main advantages of this type of algorithm are linked to the possibility to obtain a field of displacement even in the case of very low-resolution images and in extremely crowded conditions. As a matter of fact, similar processing techniques are used also in par- ticle image velocimetry (PIV) to estimate the displacement of particles in a given flow with very good performances in the case of strongly seeded media. 15 Another prerogative of digital image correlation (DIC) is that, as better explained later, the field of displacements is estimated on a regular interrogation grid in the image. In this way, it is possible to find an immediate correspondence between the estimated motion and the people belonging to an image region. In this paper, the capability of the DIC analysis to estimate the motion of jumping people is carefully qualified, mainly focusing on the uncertainty evaluation of the estimated movement as a function of the image acquisition conditions doi: 10.1111/j.1747-1567.2009.00574.x  2009, Society for Experimental Mechanics May/June 2011 EXPERIMENTAL TECHNIQUES 71