Step 3.1: Implementation - Identifying Regions with High Human Activity Cont...

The steps involved in Identifying Regions with High Human Activity are

1. Identify nodes and assign probabilities
2. Sample start and end nodes based on the probabilities
3. Simulate trajectories from the starting node to the end node
4. Calculate occupancy map from the trajectories
5. Cluster regions based on their occupancy

To summarize, given the geometry of an infrastructure, the nodes are assigned probabilities as described in the earlier steps. Later the start and the end nodes are sampled as described in 3. Given the start and end, human motion trajectories are generated as described in the previous step. These tools provide a way to simulate an entire scenario in the infrastructure. 

Calculate occupancy map from the trajectories:

By simulating the previous steps multiple times and observing the trajectories provides a way to identify regions that have high human activity. 

Figure 1.

Figure 1 shows the occupancy map by simulating the 500 trajectories as described in the previous steps. 

Cluster regions based on their occupancy:

The next step is to cluster regions with high human activity. In the step, the regions that belong to the same cluster should have a high value of occupancy and also be located in the same spacial location. The feature set representing any point are the spatial co-ordinates and their occupancy i.e. 

(x, y, z, o), where x,y,z are the 3D co-ordinates of the points and o the occupancy of the points. Expectation Maximization was used for cluster and the results are as shown below. 

Figure 2.

The mean and the standard deviation of the obtained clusters are.

Cluster	X	Z	O
Red	Mean:1784.91 Var:502.98	Mean:426.34 Var:48.08	Mean:0.0029 Var:0.0029
Blue	Mean:227.28 Var:25.95	Mean:2627.19 Var:252.04	Mean:0.002 Var:0.0016
Pink	Mean:725.58 Var:280.34	Mean:219.67 Var:25.62	Mean:0.0011 Var:0.0009
Green	Mean:2617.26 Var:258.58	Mean:633.89 Var:31.24	Mean:0.0011 Var:0.0008
Aqua	Mean:2617.2649 Var:258.5819	Mean:633.8928 Var:31.2469	Mean:0.001 Var:0.0008
Light Pink	Mean:1577.61 Var:1017.78	Mean:420.17 Var:163.06	Mean:0 Var:0

In this scenario red cluster is identified to have the highest human activity followed by blue and pink.

The next step is to find camera calibration parameters that maximizes the view of these clusters and also provides maximal frontal view of the humans.