Ground target recognition using support vector machine for UAVs

Abstract

UAVs have been powerful tools for intelligent, surveillance and reconnaissance (ISR) mission. Essential step of ISR mission is object recognition. In general, object recognition al-gorithms have heavy computational load. Therefore, they are hard to be implemented for real-time applications. The aim of this article is to address ground target recognition on aerial im-ages from UAV. In this research, two techniques to reduce the computational load will be pre-sented. First, boosted classifier with cascade structure is utilized in candidate detection proce-dure. Weak classifiers are boosted with ada-boosting algorithm. With boosting procedure, classifies are trained to extract objects correctly. Cascade structure is constructed with boosted classifiers. With cascade structure, computation time can be reduced about 10 times than the time that only boosted classifiers are utilized. Second, scale selection technique is utilized. Searching scale can be estimated with state measurements of UAV. Classification with support vector machine (SVM) is utilized for object recognition. Simulations are conducted to confirm the performance of proposed methods.