Investigating and Modelling the Microscopic Pedestrian Behavioral Dynamics in Mass Religious Gatherings

Abstract

Pedestrian models that can realistically reproduce observed pedestrian movement dynamics are crucial in facility planning and pre-event planning of crowded situations. Despite the elaborate arrangements, crowd disasters often happen in mass gatherings. It has been noted that crowd turbulence caused by multiple interactions and local disturbances can result in crowd disasters. It is in this context that understanding pedestrian behaviour at the individual level becomes crucial in ensuring the safety of the crowd. Pedestrian behaviour in a crowd is a complex multidimensional choice problem, with decisions made regarding where to go, which route to choose, which activity to perform, how to sequence the activities, whether to stop/move, which direction to choose and many more. This is also attributed to the continuous interaction that happens with other pedestrians and the surrounding environment and the larger degree of freedom available to pedestrians. Models that can capture this complex behaviour to a reasonable extent can be used to aid event planners in deciding crowd management strategies. It could be said that there is an array of walking manoeuvres used by pedestrians to facilitate their movement in a crowd, where specific manoeuvres might be more suited than others depending on the situation at hand. Among the plethora of walking strategies that a pedestrian might choose from, this thesis tries to identify an order of preference of walking manoeuvres used in representative crowd situations that will help us gain more insights into walking dynamics and thereby make better predictions.

Development of realistic pedestrian models requires gaining more insights into pedestrian behaviour through observation of their walk dynamics in natural environments. For this purpose, empirical studies are conducted using video data collected on pedestrian movement in mass religious gatherings in Ujjain. Controlled laboratory experiments were also carried out to capture specific pedestrian behaviour and compare the differences in pedestrian behaviour across different data collection techniques.

As a preliminary step, the effect of the geometry, different flow conditions, sociodemographic characteristics, and the effect of social groups on walking dynamics are studied. The highest mean walking speeds of about 1.11m/s are noted on open roadways without barricading. The impact of socio-demographic factors on pedestrian walking speed is carried out through appropriate statistical tests. The test would help us identify differences in the mean speed of pedestrians across different groups obtained by social categorization. A comparison of the mean speed of pedestrians in the field vs experimental setups indicates that the surrounding environment has a significant influence on the mean speed of pedestrians. Further, comfort while walking is also a function of personal space around an individual. It varies from person to person depending on culture and context and there are significant individual differences too. This study tries to formulate and define personal space in crowded situations. Video data of pilgrims taking part in the Panchkroshi Yatra, a religious walkathon, is used for the analysis of factors affecting personal space. It is also observed that personal space follows an asymmetrical pattern rather than a symmetrical pattern. Pedestrians usually walk in social groups, and these group interactions significantly affect their behaviour. We further study the effect of groups on walk dynamics by performing a series of controlled experiments. The pedestrian speeds were found to be 1.09±0.087 m/s for single pedestrians, 0.92±0.047 m/s for pedestrians in group size 2, and 0.86±0.045 m/s for pedestrians in group size 3. Often group members act in response to the perceived changes, sometimes by taking responsibility for each other and waiting for others to catch up, or at other times it could be in terms of speeding up and moving ahead. We investigated these interactions by exploring the changes in the spatial arrangement/formation over time.

Next, the impact of specific activities and activity locations are studied as a function of accessibility to activities on the tactical level decisions using visual analysis and spatial analysis tools. One of the primary activities for pilgrims in most mass religious gatherings involves performing certain rituals. In Kumbh Mela, taking a dip in the river (at riverbanks known as “ghats”) is considered to be extremely auspicious. For this reason, we try to evaluate the microscopic parameters such as speed variations, and local densities at the ghats. It is also observed that the duration of the dip and the visual perception of crowding impact the accessibility to that activity. Further, we explore the role of accessibility of prime activity locations and the critical links in the context of crowd management. This is carried out by using the theory of space syntax to arrive at indicators for visual accessibility.

An interesting phenomenon that occurs in such mass gatherings is the interception of a pedestrian stream by a small segment of pedestrians. This trickle of pedestrians disrupts the flow and creates local disturbances. Here, we explore this crossing behaviour of pedestrians in the event of conflicts in the direction of motion. A new way to model this behaviour is proposed by formulating the decision-making mechanism used by pedestrians to overcome conflicts as gap-seeking behaviour. Pedestrians looking for gaps or spaces in a crowd to facilitate their movement form the basis for such an analysis. Multiple gap choice decisions of individuals are modelled to understand the effect of individual-level heterogeneity on gap choices. The analysis results also indicate that pedestrians tend to force gaps to facilitate movement in their natural state. Consequently, controlled experiments cannot reproduce or motivate the participants to behave as in a real crowd.

The following (or) queuing behaviour of pedestrians, wherein people walk one after the other in a line is a common occurrence in pedestrian facilities or crowd gatherings. Though many studies have explored this behaviour, especially using single-file experiments, this behaviour cannot be treated in isolation from other walking behaviours like overtaking or lane-changing. A methodology to model the pedestrian walking dynamics in a unidirectional single-channel flow is proposed by jointly modelling the aspect of probabilistic consideration of leader, lane changing, and following behaviour. It is noted that the probability of lane change is directly proportional to the adjacent lane spacing and inversely related to the current lane spacing. The threshold spacing, within which the chances of the following pedestrian to consider the pedestrian ahead as a leader is obtained to be 1.2 m for the final model. An agent-based model is developed, where the rule set is derived from the calibrated behavioural model. It was observed that when few pedestrians stopped in between for a while, the average speeds reduced from 1 m/s to 0.57 m/s and the probability of following pedestrians doing lane changes increased. The predictive ability of the model in reproducing the logical behaviour of pedestrians is demonstrated using what-if scenario analysis considering speed variations, and flow variations. A pedestrian walking model that can account for more naturalistic behaviour is also developed for unconstrained flow situations. This requires us to treat walking not just as the means for navigation, but navigation where people look around and often wait for their group members or pause to watch a road-side activity. Several other factors contribute to these decisions and make walking a ‘pause-and-go’ behaviour with occasional pauses in walking. This intermittent walking behaviour is modelled using a bi-level modelling framework with the decision to pause at the higher level and the choice of direction at the lower level. The proposed model is calibrated using real data of pedestrian movement. The model is further validated using another dataset and the average percentage of correct predictions for an individual is about 70%.

Summarizing, this thesis attempts to identify the decisions made by people while walking, the factors influencing these decisions, and model some representative crowd situations by postulating the decision-making process of an individual in a specific context. It is hoped that the proposed models and the associated inferences could be used to improve the behavioural realism of crowd simulations and thereby make better predictions.