A system to automatically track humans and vehicles with a PTZ camera

Uses a pan–tilt–zoom (PTZ) camera to track vehicles and pedestrians. For background modeling it assumes static background (assumes moving objects will have incoherent motion and can be eliminated with post-processing filtering) and uses a variant of probabilistic neural network (PNN) and it allows producing a background probability map. Groups of “connected” pixels in the background probability map form blobs.

Tracked objects are characterized by their shape, color, motion, class label. Objects are classified to vehicle/pedestrian/unknown category based on their geometrical properties. It checks for certain features to detect pedestrians, measured classification error was ~3%.

It can also use PTZ cameras to track certain objects, using the color signature of the object. For this, a particle filtering approach ( #142 ) is used based on color histogram information. Because of the many histogram comparison, for efficiency it uses a method called "Integral histograms".

The system is said to have achieved fairly good results under different conditions.

kundralaci / JAMResearch

A system to automatically track humans and vehicles with a PTZ camera #53