Tracking Objects Based on Multiple Particle Filters for Multipart Combined Moving Directions Information

Algorithm 4

Sample to train gentle Adaboost.

	Input: Image at initial time, position and object size
	Output: Data set D size , including positive patern, negative patern.
	Step 1: Initialize a feature stack to hold feature vectors.
	Step 2: Get the Haar-like characteristics [34] of the area ,
	Put on stack features, features.push ()
	Step 3:
	for i = 1 to N+ −1 do
	Beginfor
	Randomly draw rectangular area S from the image at a position of ±5 pixels from the object
	Get the Haarlike feature vector over the region S, v = getHaarlike(S)
	Put vector in the stack features, features.push(v)
	Endfor
	Step 4:
	for i = 1 to N−−1 do
	Beginfor
	Randomly take the area of S rectangle in the green part of Figure 8
	Get the Haarlike feature vector over the region S,  = getHaarlike(S).
	Put vector in the stack features, features.push ()
	endfor