Tuesday, November 11, 2014

Step 1: Motivation and Idea

Motivation
Surveillance is an integral part of many public area infrastructures like airports, banks and train stations. Networked camera systems are the most common mode of surveillance. Given the infrastructure or the geometry of the environment, the positioning and orientation of the cameras can play a significant role in constructing an effective surveillance scenario. The definition of effective surveillance is subjective and is primarily defined by the specific scenario. For example, the deployment of cameras in a public area like a movie theater is more relaxed compared to an airport. In a movie theater it might be sufficient to deploy cameras at locations that exhibit large human activity, but in an airport it is imperative to deploy cameras such that a maximum visibility coverage can be obtained irrespective of the amount of human activity in the area. Despite this, some common factors have to be taken into consideration before deploying the cameras.

Visibility Coverage: In high security scenarios, the camera configuration should be optimized such that a maximal coverage of the infrastructure can be obtained. In low security scenarios, the camera configuration should at least guarantee the coverage of all the areas that have large human volume. The configuration should also guarantee the coverage of the most frequently used entry and exit points in the infrastructure. Furthermore, a camera configuration that captures the frontal image of the humans as opposed to their posterior images is more effective. 

Deployment Cost: On the other hand it is important to take into consideration the cost effectiveness of a camera deployment configuration. The configuration should guarantee maximal coverage while deploying the least number of cameras. Furthermore, having a minimal required number of  cameras has a significant impact on the available storage space. Eliminating a single redundant camera can save storage space up to 720 hours of video every month. HD cameras are becoming more prevalent these days and require higher storage space.

Given the geometry of the infrastructure, designing a camera deployment configuration manually by taking into consideration the above factors can be extremely tedious and error prone. Automated camera network deployment optimization techniques are very essential for a cost effective and safe environment. Furthermore, it is unusual to consider the configuration of the camera network during the design and construction of an infrastructure. Factoring this in during the design planning can be very cost effective. It can help in making design decision that result in a safer environment by avoiding areas where the deployment of cameras may be hard. Most camera networks require cabling, knowing the location of the cameras a head of time can decrease any excess cost associated with it.

Idea
The problem is that, given the 3D geometry of the infrastructure, what configuration of camera deployment will provide effective surveillance. As discussed before, the idea of  effective surveillance is subjective. Though a scenario that has maximal visibility coverage can be considered to be effective, such problems have been explored before as a more prominent Art Gallery Problem [1] which are considered NP-hard. In this work, a network configuration is considered to provides effective surveillance if,
  • it provides coverage of the most common entries and exits in the geometry.
  • it provides coverage of all the areas in the geometry that have high volume of human activity.
  • it provides a maximal amount of frontal view of the person as possible instead of their posterior.
Identifying the prominent entrances/exits: All public infrastructures have entrances, exits and points of interest. Any doorway can be considered as an entrance or an exit, for simplicity they are referred to as a node. In an infrastructure different nodes are accessed with different frequencies. A node representing a common entrance or an exit has a high frequency of access as opposed to an employees personal office. So the first step would be to identify the nodes in an infrastructure along with their probability of access. 

Identifying areas with high volume of human activity: Areas with high volume of human activity can be identified by simulating the trajectories humans would follow in an infrastructure and locating the areas that have high volume of access. In general Humans traverse hallways with the objective of reaching a destination or a goal (unless loitering). Assuming that the trajectories within the infrastructure are generated by humans trying to reach between different nodes, and as the probability of the nodes were already estimated, the start and the end nodes can be sampled according to the probabilities and a trajectory can be simulated from the beginning to the end. By simulating the trajectories multiple times, areas of high access can be identified. 

Identifying orientation to maximize frontal view: The simulated trajectories in the previous step also has information regarding the direction of motion. The idea is to choose a camera location and orientation such that the camera points in a direction that is opposite to direction of motion of most simulated trajectories. 

To summarize, the idea is to identify camera locations and orientations in the infrastructure that optimizes the above constraints.


References

No comments:

Post a Comment