Video Pedestrian Detection Based on Orthogonal Scene Motion Pattern

Algorithm 2

Look-up-table.

 for  each weak learner ℎ∈𝐻  do

  Partition feature space into 𝐽 disjoint blocks 𝑥₁,…,𝑥_𝐽 that at least one sample in each block;

  Under the distribution 𝐷_𝑡 calculate 𝑊^𝑗_𝑙=𝑝𝑦_𝑖=𝑙∣𝑥_𝑖∈𝑋_𝑗 where 𝑙=±1;

  Calculate 𝑍_ℎ=∑_𝑖{𝐷_𝑖exp[−(1/2)𝑦_𝑖∑_𝑗(𝑊^𝑗₊₁−𝑊^𝑗₋₁)𝐵_𝑗(𝑥_𝑖)]} where 𝐵_𝑗(𝑥)=1,𝑥∈𝑋_𝑗;0,𝑥∉𝑋_𝑗;

 end for

Output:  𝑓_𝑡(𝑥)=^defarg_ℎmin𝑍_ℎ.