- project with waqas sultani

	- measure crowd density using optical flow

		why?
		- object detection on humans/heads becomes too difficult to track.
		- optical flow is easy to compute on crowds and is generally smooth.

		how?
		- crowd can be treated as particles in the flow field (particle advection).
		- particles represent relality better as measurable discrete objects rather than continuous flow fields.
			- therefore, density can be estimated more easily with these objects rather than continuous flow fields.
			- also, computations are done where the mass is located, and not throughout the entire continuous field.
		- three algorithms:
			- (1) social force using Lagrangian specification of the flow field (track particles using particle advection).
			- (2) Eulerian specification of the flow field (observe change in flow at fixed locations).
  			- (3) Delaunay Tessellation Field Estimator or Smoothed Particle Hydrodynamics (density estimation in particle hydrodynamics, usually used for molecules and astrophysics).
		- by using the three algorithms, it is possible to train an AI that can interpret these information to formulate an estimated crowd density.
		- ultimately, the goal is to provide an accurate number of people in a sample video.

		advantages and disadvantages of each algorithm:
		- (1) great for tracking particles. simulates "freezing by heating". information about density can be extracted from the force flow of the particles. tracking movement of objects can tell us they go around dense areas (don't know if its a wall or a group of people though).
		- (2) can tell us that a change in velocity occured (possibly signaling that a dense object has formed.  Also, if the velocity never changes, it is possible there is a wall there).
		- (3) measures density based on the positions of the particles. using the information provided by (1) and (2), one can possibly estimate the density very well.

		- (1) tells you that you are moving around something dense.
		- (2) can possibly tell you if the object they're moving around is a stationary object, or dense group of people.
		- (3) can possibly measure the density by using (1) and (2).

	- papers i have read in order of decreasing understanding.

		Crowds:
		- Abnormal Crowd Behavior Detection using Social Force Model
		- Abnormal Traffic Detection
		- Many Particle System
		- DTFE
		- SPH
		- crowd density estimation based on statistical analysis of local intra-crowd motions for public area surveillance
		- counting crowded moving objects
		- A Streakline Representation of Flow in Crowded Scenes
		- Crowd Density Estimation Using Texture Analysis and Learning

		Optical Flow:
		- High Accuracy Optical Flow Estimation Based on a Theory for Warping
		- A look at motion in the frequency domain
		- optical flow using spatiotemporal filters

	- Things i (have done recently)/(am doing):
		
		- Currently coding up Abnormal Crowd Behavior Detection using Social Force Model.
			- finished particle advection
			- attained very reliable optical flow algorithm

		- Coded up a decent, but not heavily tested, algorithm on density estimation on my own.
