Since the fatalities accidents in building fires occur more frequently and the safety issues of the lager-scale crowds that gather in a place for entertainment, ceremonies or education becomes an important element, more and more attention has been paid to understand human behavior in emergency and occupant evacuation in buildings, furthermore, considerable evacuation models and software have been developed. These models are mainly used to reproduce occupant evacuation process and predict evacuation results (such as evacuation time and occupant flow rate) in specific conditions. However, the calibration and validation of the models need lots of convincing empirical data. Unfortunately, the reliable data and information with regard to pedestrian and evacuation dynamics is still scarce. Additionally, there are few models focusing on reproducing and exploring the specific human behavior during evacuation process, such as selection behavior.
In this paper, we will study carefully the human choice behavior including exit selection and aisle selection based on the controlled evacuation experiments. We have conducted a series evacuation experiments. There were 102 college students took part in the experiments as evacuees. The place carried out the experiments was a university building and the experimental area consisted of a classroom, a passage and two stairs. During the experiments, at each end of the passage, there was an alarm which can give audible information representing stair exit status (open or closed). What needs to be emphasized is that the information may be false alarm or not activated which we controlled intentionally as a variable to study the respond of people to audile information during evacuations. The experiment variables consisted of alarm status, status of the stair exit, and the type of occupant initial distribution. There were two types of occupant initial distribution. One is nonsymmetrical distribution, that is, there were more individuals in the back of the classroom than that in the front of the classroom. And the other is symmetrical distribution, namely, all the students distributed in the classroom nearly evenly. We positioned totally nine cameras to record the experiment process.
By controlling the experimental variables, we conducted totally eight experiments which were divided into two categories: without and with particular groups. For without particular groups situations, all individuals had the same instructions, at the beginning of the experiment, all students sat on their own position in the classroom, and started to evacuate as soon as possible when the audile start signal was given. For each evacuee, the evacuation fell into three successive phases. Firstly, he/she had to move out of the classroom, and then chose a movement direction in the passage, moved down to the end of the passage and then went downstairs to the destination which was under the next below floor. If someone moved to a closed stair exit, he/she had to move back and evacuated via the other stair exit to the destination. All participants wore red hat in order to indentify clearly. For with particular groups situations, we chose randomly several students as particular individuals, and they were told that they
had to meet together at the area near the classroom exit in the passage before they moved to stair exit; furthermore, they had to keep together during the whole evacuation process. The ordinary individuals were the same as those in the first category, who were just told to evacuate to destination as fast as possible. The ordinary individuals wore red hats, and the particular individuals had different color hats according to different group. In addition, in the second category of experiments, there was only one stair exit (stair B) available, and it was told to all before evacuation, so they need not to estimate which escape direction was available when they arrived at the passage. The students were very familiar with the building, and highly motivated. During the experiment process, they were asked moving as fast as possible. After each experiment, all students came back to the classroom, and there was five minutes to rest, so they did not represent fatigue during the experiments. If anyone felt tired, he/she can drop out before each experiment. At each experiment, each individual sit on different initial position.
Based on the controlled experiments, we find that it is non-symmetrical for pedestrians’ exit selection and aisle selection in classroom. In the first category of experiments, even though there were more individuals situated in the back of the classroom, the choice percentage of classroom front exit A was larger than that of classroom back exit B. This deviation was more apparent for the second category of experiments. Differently, there were much more individuals chose the classroom exit B. A massive number of pedestrians chose the middle aisles which were between seats, however, only few people chose the aisles which were next to the wall. In addition, bifurcation point, where people located at the same row divided into opposite direction, was biased towards the classroom exits.
We found that pedestrian destinations, that is, the choice of stair exit had significantly effects on occupants’ choice of classroom exit which was pedestrians’ intermediate destination, but the initial occupant distribution had little influence; pedestrians located in middle area tended to choose the aisle which is near the exit at the cost of taking more lateral movement between seats.
To explore the mechanism of the non-symmetrical choice behavior, we will simulate the evacuation process based on our proposed cellular automata model [1-2] and a microscopic simulation tool called NOMAD  which was developed in the Transport & Planning department of the Delft University of Technology. In the study, we will reproduce the non-symmetrical choice behavior, and then we will compare the results between experiments and simulations by considering the factors we discussed above (initial occupant distribution, expected destination, psychology behavior) to find which factors is most important.
The results in this paper are expected to provide valuable advice for large-scale crowd management both in normal and emergency situations and the optimization design of infrastructure layout in multi-obstacles buildings (such as classroom, theatre, and stadium).