Stages in the cascade are constructed by training classifiers using Gentle AdaBoost. A positive result from earlier strong
classifier triggers next strong classifier which has also been adjusted to achieve higher detection rate than previous one. A negative result is immediately rejected at any stage of cascade structure.