A TS Fuzzy System Learned Through a Support Vector Machine in Principal Component Space for Real-Time Object Detection

This paper proposes a Takagi-Sugeno (TS) fuzzy system learned through a support vector machine (SVM) in principal component space (TFS-SVMPC) for real-time object detection. The antecedent part of the TFS-SVMPC classifier is generated using an algorithm that is similar to fuzzy clustering. The dimension of the free parameter vector in the TS consequent part of the TFS-SVMPC is first reduced by principal component analysis (PCA). A linear SVM is then used to tune the subsequent parameters in the principal component space to give the system better generalization performance. The TFS-SVMPC is used as a classifier in a camera-based real-time object detection system. The object detection system consists of two stages. The first stage uses a color histogram of the global color appearance of an object as a detection feature for a TFS-SVMPC classifier. In particular, an efficient method for histogram extraction during the image scanning process is proposed for real-time implementation. The second stage uses the geometry-dependent local color appearance as a color feature for another TFS-SVMPC classifier. Comparisons with other types of classifiers and detection methods for the detection of different objects verify the performance of the proposed TFS-SVMPC-based detection method.