An Extended Set of Haar-like Features for Bird Detection Based on AdaBoost

Abstract

As Adaboost is an efficient method to select features, we investigate its capability in selecting effective features from a set of Haar-like features for bird detection. Moreover, Adaboost is also used to construct a strong classifier for the task of detecting birds from a set of weak classifiers. We propose to add specially designed new Haar-like features to increase the detection rate of interested objects, here, namely birds. Our experiment shows that this method can increase the TPR (True Positive Rate) and decrease the FPR (False Positive Rate). In addition, a motion detection algorithm is used to detect moving objects, which are segmented from background. Using the strong classifier trained from Adaboost algorithm, segmented objects are classified as bird or not. The advantage of this approach is that higher performance and real-time detection can be achieved due to the fact that the strong classifier needs not to examine all the possible sub-windows of the input image. Thus, this method can further decrease the rate of false positive.