Video Pedestrian Detection Based on Orthogonal Scene Motion Pattern

Algorithm 1

Gentle AdaBoost.

Given:  (𝑥₁,𝑦₁),…,(𝑥_𝑛,𝑦_𝑛),𝑥_𝑖∈𝑋,𝑦_𝑖∈{−1,+1};

Initialize:  𝐷_𝑖=1/𝑛,𝑖=1,2,…,𝑛,𝐹(𝑥)=0;

 for  𝑡=1,2,…,𝑇  do

  Fit the regression function 𝑓_𝑡(𝑥) by weighted least-squares of 𝑦_𝑖 to 𝑥_𝑖 with weights 𝐷_𝑖;

  Set 𝐹(𝑥)←𝐹(𝑥)+𝑓_𝑡(𝑥);

  Set 𝐷_𝑖←𝐷_𝑖exp[−𝑦_𝑖𝑓_𝑡(𝑥_𝑖)],𝑖=1,2,…,𝑛, and normalize 𝐷_𝑖 so that ∑𝐷_𝑖=1;

 end for

Output: the classifier sign[𝐹(𝑥)]=sign[∑𝑓_𝑡(𝑥)+𝑏] where 𝑏 is threshold.