The Pedestrian Detection System Based On HOG And STM

Pedestrian detection has always been a hot topic in the field of computer vision research.By detecting,analyzing and tracking pedestrian in images or video to automatic realize management and monitoring.Pedestrian detection has high commercial prospects and practical value in practical applications such as automobile safety,visual monitoring,and automatic driving.Currently,HOG+SVM has made a great achievement in pedestrian detection,providing a broad idea for later researchers.The content of this study is to improve the accuracy and recognition rate of pedestrian detection by improving related algorithms.Support vector machine is a commonly used classification method that can classify vector data.But it also exists some shortage.For example,some tensor data,such as images,its structural information inside will be destroyed when it is converted into a vector.Therefore,scholars come up with the support tensor machine algorithm and make some achievements.After years of research,tensor decomposition of CP is usually used to process the data in the support tensor machine.Image is a kind of tensor data.The HOG feature is a three-dimensional tensor.For example,the image with 64×128 size has a HOG feature of 7*15*36.When using the SVM classification algorithm,the tensor needs to be converted into a 3780-dimensional vector.But the support tensor machine algorithm can directly train the tensor data.In theory,using a tensor machine support algorithm should have more advantages.For example,for an image of 64*128,its HOG feature is a third-order tensor whose size is 7*15*36.On this basis,to improve the performance,another Tucker decomposition method is adopted to process tensor data in this paper,On the basis of the predecessors,this paper combines HOG features with STM for pedestrian detection.On the other hand,another support tensor algorithm using Tucker decomposition is proposed to use in STM and hope to a good performace.Experiments show that the support tensor machine based on the Tucker decomposition has obvious advantages over SVM in processing image data classification problems.And using this method The recognition rate is improved.Meanwhile,the HOG+STM method proposed in this paper has a better advantage than HOG+SVM.