Cascaded AdaBoost - 2022

Purpose

• Construct and testify the cascaded design of AdaBoost classifiers to achieve an arbitrarily low false positive rate. 

• (Reference: Viola Jones object detection framework.)

Viola–Jones object detection framework

• a boosted feature learning algorithm

• trained by running a modified AdaBoost algorithm on Haar feature classifiers. 

• every weak classifier is a threshold function based on the feature.

• the threshold and the polarity are determined in training.

False Positive (FP) and False Negative (FN)

Results

• The false positives decrease as expected because the goal of each cascade stage is further to eliminate false positives from the previously identified positives.

• The false negatives increase as expected because previously identified negatives are discarded and are not passed down to the next cascade stage.