Abstract
We propose a new method to optimize the completely-trained boosted cascade detector on an enforced training set. Recently, due to the accuracy and real-time characteristics of boosted cascade detectors like the Adaboost, a lot of variant algorithms have been proposed to enhance the performance given a fixed number of training data. And, most of algorithms assume that a given training set well exhibits the real world distributions of the target and non-target instances. However, this is seldom true in real situations, and thus often causes higher false-classification ratio. In this paper, to solve the optimization problem of completely trained boosted cascade detector on false-classified instances, we propose a new base hypothesis weight optimization algorithm called DOOMRED (Direct Optimization Of Margin for Rare Event Detection) using a mathematically derived error upper bound of boosting algorithms. We apply the proposed algorithm to a cascade structured frontal face detector trained by AdaBoost algorithm. Experimental results demonstrate that the proposed algorithm has competitive ability to maintain accuracy and real-time characteristic of the boosted cascade detector compared to those of other heuristic approaches while requiring reasonably small amount of optimization time.
Highlights
The boosted cascade detector [1] became the most popular method for an object detection in computer vision
We propose a fast algorithm called DOOMRED that optimizes the base hypothesis weight set of each single layer detector in a boosted cascade detector, especially when the false-rejected target instances are enforced
In a boosting algorithm, the base hypothesis selection procedure from the large candidate base hypothesis set usually demands high-computational cost. This is the reason why we focus on the optimization of the base hypothesis weight set for the performance enhancement of a boosted cascade detector
Summary
The boosted cascade detector [1] became the most popular method for an object detection in computer vision. Most researches on the boosted cascade detector have concentrated on the learning problem for a fixed number of initial training data. The basic assumption made in the researches is that the distributions of the target and nontarget objects obtained from the given fixed number of initial training data are good enough to reflect the real distributions, which is seldom true in practice. This is because it is almost impossible to know the exact distribution of the target as well as nontarget instances in real situations. The detector trained with the fixed number of initial training data cannot work properly in the real applications
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have