Experimental Results

In order to verify the system as a whole, all the corresponding components had to be brought together, and tested on varying environments.

Running a test scenario's not only allows us to see how effectively the rule-sets describe pedestrian flow, but also provide a means of evaluating how well the potential field applies to complex environments.

Each test run is for different environment dimensions, however each cell size remains consistent at 0.4m across. This value is representative of the average size a pedestrian; thus the area taken by a pedestrian is 0.16m^2.

The first environment tested, was designed to test the behaviour of people moving through a thin door. This scenario is good since it isolates a potential problem situation in most environments; that of the bottleneck at a door.

The test environment was set up as a 80 by 10 cell corridor (32 by 4m) with a partition dividing it into two. This partition has a single cell doorway that the pedestrians must move through. The environment is seeded with 100 randomly distributed people, all of whom are moving to a line of attractors situated to the right of the environment.

The empty hall test environment and the corresponding potential field. The pedestrians flow from left to right

The experiment was run under identical starting conditions, but changing the rule-set each time, over Ruleset 1, Ruleset 2 and Ruleset 3. Results for this experiment are shown below, where each image was taken at 15 step intervals (the time for one step is not yet converted into seconds, this would require calibration against real evacuation times).

The first rule-set tested, Ruleset 1, provided a queuing behaviour for the pedestrians. Interestingly, as simple as the rule-set is, it still can be considered in the context of realistic pedestrian phenomena. Similar behaviour to this may be seen in an orderly evacuation procedure, or even people leaving a concert hall after a performance. The ordered queuing may initially not visually match the `evacuation' flow, however, considered in the correct context it can be seen to match orderly flow near perfectly.

Here the pedestrians are moving down the hall to the line of exits on the right, and each frame is taken at 15 step intervals. The movement is being described by Ruleset 1, and describes a `queueing' behaviour similar to what one might expect to see of people leaving a concert hall.

The second rule-set tested involved considering diagonals after forward checking for occupancy. As the results show, this breaks down the linear queuing that was evident in Ruleset 1. The result is a large oval shaped congregation of people at the door. This scenario visually resembles a less ordered evacuation of a building, or a less considerate crowd leaving a venue. There is an apparent lack of queuing as it is evident that people are pushing past others in order to reach the exit.

The movement shown here is described by Ruleset 2. This rule provides ordered evacuation behaviour where queueing is still visible, however the line is far less ordered than described by Ruleset 1.

Ruleset 3

The last rule-set tested, Ruleset 3 included horizontal direction checking after diagonals and forward checking. The results, aren't radically different from Ruleset 2, however the shape of the congregation has changed from a pointed oval into a flattened circle against the wall. This scenario is more indicative of a panic scenario, where there is visible disorder in the crowd and the resulting flow is more turbulent as a result.

The movement described by Ruleset 3 provides a more frantic evacuation model. The crowd is less ordered and there is a visible crush of people at the door trying to push themselves into the optimal position.

Simple Office Floorplan

As a means of putting all the results together to demonstrate the workings of the simulation in a complex environment, I have chosen to test the flow of people out of a synthetic office space. The office space contains only a single exit located in the top left of the map and has dimensions, 60 by 60 which corresponds to a floor space of 576m squared. Over this floorspace 200 people have been randomly distributed.

The potential field algorithm copes nicely with this level of complexity as is visible in the contour map below, height bands are visible flowing and refracting through the openings. The 3d potential field for this environment is shown in the routing page.

The pedestrian rule-set chosen for the evacuation simulation was Ruleset 3, the reason being that behaviour of a panicking crowd would aid in the identification of any potential bottlenecks caused by the design of the Office Layout; results are shown below where each image was taken at a 10 step interval.

Initial Results

The empty office floor and the corresponding potential field. The potential field has been thresholded to aid its visualisation

The results of the pedestrian flow provide an excellent overview of pedestrian evacuation in the building, and congestion is immediately apparent in the top left of the map. Since this software provides a simple means to alter the building design and visually see the changes, we can now attempt to alleviate the bottleneck occuring in the top left of the map. Since we are funneling people into a tight space, we can move the exit lower into the more open area on the south wall of the exit room as is shown in the next example.

Pedestrian flow according to ruleset 3 is shown with 10 time steps between each frame. Frames are ordered from top left down to bottom right, and the exit is located in the top left corner of the map.

Moved Exit

Here the problem exit has been moved to a more open area with the hope of reducing the bottleneck and decreasing total evacuation time.

The pedestrian flow resulting from the rellocated exit are suprising in the sense that instead of reducing the bottleneck, they introduce a bigger one. On closer inspection of the results, we can see that moving the exit lower provides a shorter exit route for other pedestrians lower on the map, and instead of taking the corridor on the left, they opt for the new route. This new bottleneck actually increases overall evacuation time.

Here one of the crucial evacuation exits has been moved lower. The results show an increased bottleneck, and consequently slower evacuation times. All other parameters are the same as the previous example.

Extra Wall

A wall has been added to the previous map opposite the initial bottleneck location. This is to show what effect environement changes have upon pedestrian flow.

The final set of results I have included, shows how the addition of walls into an evironment can radically alter pedestrian flow, and evacuation time. For the final map I have added a wall blocking off access for pedestrians to the east of the map. This limits the pedestrian to flow clockwise around the centre room. Results show the increased evacuation time.

A wall has been added to the room where the right part of the map evacuate through. This has forced a different, longer route to be used, increasing evacuation time. Again all other parameters are identical as before.