Process and ImplementationOptical FlowFirst, I had to study the methods to calculate optical flow. From the description in the paper, it looked like they were using the Lucas-Kanade method. This method assumes that the optical flow is constant in a small window. This is a widely used method of calculating optical flow and is found in several libraries. The library I used was created by Piotr Dollar at the University of California, San Diego (1). This toolbox also implements a cross-correlation method of calculating optical flow, as well as the Horn & Schunck method. I tested these three methods and determined that the Lucas-Kanade method was the best choice for this application. In class, it was suggested that we look at a hierarchical version of the Lucas-Kanade method to more accurately calculate optical flow, as it is an integral part of this paper. I tested a toolkit (2) written by Sohaib Khan, who was a student of Mubarak Shah at the University of Central Florida. This code produced much more noise-sensitive results, and I determined that the best choice was the original toolkit. These results can be seen in Figure 1. Figure 1: Lucas-Kanade optical flow method (left) versus hierarchical Lucas-Kanade optical flow method (right) Generalized social force for particle advection and calculation of the interaction force. The article presents a method of using particle advection to estimate the flow of objects through a scene. Since crowd scenes are cluttered and difficult to follow under a microscope, the authors proposed this macroscopic method of superimposing a grid of particles on the scene and advecting them via optical flow to model crowd movement. This... middle of paper ... is unclear and remains a mystery to me. I would also be interested in implementing the concept other than particle advection, as I feel it loses a lot of data by not evenly covering the scene in each clip. Dollar, Piotr. Piotr's Matlab Image and Video toolbox. Piotr Dollar. [Online] February 22, 2010. [Cited: March 17, 2010.] http://vision.ucsd.edu/~pdollar/toolbox/doc/.2. Khan, Sohaib. Source code. Mubarak Shah. [Online] [Cited: March 17, 2010.] http://www.cs.ucf.edu/~vision/Code/Optical_Flow/Lucas%20Kanade.zip.3. Blei, David. Latent Dirichlet allocation in C. David Blei. [Online] [Cited: March 17, 2010.] http://www.cs.princeton.edu/~blei/lda-c/index.html.4. Mochihashi, Daichi. lda, a latent Dirichlet allocation package. Daichi Mochihashi. [Online] December 4, 2004. [Cited: March 17, 2010.] http://chasen.org/~daiti-m/dist/lda/.