基于支持向量约简的快速目标检测

支持向量机(SVM)由于其出色的泛化能力,已成为目标检测领域应用最为广泛的分类器之一.然而在检测过程中,过多的支持向量会产生很大的时间开销,从而降低目标检测系统的实时性.针对此问题,提出一种约简支持向量的方法,以降低分类器的决策开销,加快检测速度.此方法采用迭代的方式来估计特征空间中向量的原像,通过构建精简原像集来简化支持向量机,从而达到了提升分类速度的效果.利用精简的SVM结合Selective Search+ BoW模型构建了一款快速检测器,测试结果表明:该检测器能够在保证检测率的前提下,通过约简支持向量,提高目标检测的实时性.

Fast Object Detection Based on Support Vector Reduction

The support vector machine (SVM) has become one of the most widely used classifier in object detection due to its excellent generalization ability. However, during the process of object detection, too many support vectors will cause huge time cost, which degrades the real-time performance of the object detection system. To solve this problem, a support vector reduction method is proposed to speed up the detection speed and reduce the decision-making cost of classifier. This method iteratively estimates the pre-images of the feature vector in the feature space, and the reduced pre-image set is applied to construct a simplified SVM, which achieves fast speed in object detection. In this paper, an ensemble detector which combines simplified SVM, Selective Search and BoW (Bag of Words) is proposed. From the test results, our detector reveals fast detection speed and guarantees high detection rate by reducing support vectors.