This project is a continuation of the traffic car counting.
Assume that the cars are represented on a graph that is continuously
changed, a stream graph.  In the graph, the vision algorithms can give us
with high enough confidence the color of the car, the blob size, an approximation
of its speed, and the relative position
of the center of their blob mass on the background.  With this we create a 
graph for each frame obtained from the vision program.
The problem in its simplest form is given two consecutive frame graphs,
develop  a subgraph isomorphism algorithm that takes into account
the uncertainties and positional semantics of solid objects
to guarrantee high enough accuracy
in identifying the cars from frame to frame. 
The issues of uncertainty that need to be dealt are small variations on the color, 
the change of the blob size changes as cars move toward or away from the camera, 
and relative positions in the two frames.
This is to be done as an online algorithm that takes 1-2 frames per second and
continuously identifies the cars in a 2-6 sequence of frames as they cross the viewfield
of the camera.

The next step is to solve the problem for image frames coming from
different cameras. Depending on the distance between cameras,
the relative positions could change drastically, and more contextual
knowledge must be used to make reasonable identifications.