A study of object detection based on fuzzy support vector machine and template matching

Fuzzy support vector machine (FSVM) is a novel type of learning machine, it has shown to provide better generalization performance than traditional techniques. This thesis introduces the theory of FSVM briefly and application in an object detection system, and discusses in detail the core techniques and algorithms, which combine FSVM and template matching into two-layer serial classifier. The algorithm takes the object detection problem as two classes classification, thus it can take the advantages of the Fuzzy SVM. The algorithm takes template matching as the coarse filter, which can reduce the amount of data and getting rid of redundancy of the data sets. Template matching can also diminish the difficulty of SVM training, and speed up the training progress. Moreover, the Normalized correlation coefficients are computed in the stage of template matching. These coefficients are used to determine the fuzzy membership without any additional computation. The experiments show that the algorithms have low error rate and high speed.