Imaging sonar based real-time underwater object detection utilizing AdaBoost method

We propose a real-time underwater object detection algorithm using forward-looking imaging sonar. Considering the characteristics of sonar image, the Haar-like feature is used to construct each weak classifier. We construct a strong classifier by combining several weak classifiers. An adaptive Boosting (AdaBoost) algorithm is utilized to determine coefficients of each weak classifier and weights of training dataset. Moreover, we improve the efficiency of calculation using a cascade structure. To verify our method, we use the field data obtained by hovering-type AUV “Cyclops”. From this data, we create a training dataset and conduct the learning process of detector. The experiment results show the accuracy and tolerance of the object detector made by the proposed approach.