A Fast Pedestrian Detection Algorithm Based On Deformable Component Model

In recent years,with the development of computer vision technology in China,more and more attention has been paid to the application of computer vision technology.As the most basic work,pedestrian detection is the most difficult and basic work in the field of graphic processing and computer vision analysis.The problem and technology of pedestrian detection has gradually become a hot research trend of computer researchers in recent years.Moreover,with the continuous development of digital or video image processing technology at home and abroad,Pedestrian detection technology has been applied to all aspects of people's lives,such as traffic violations detection;in medical treatment Therefore,the pedestrian detection system is very strict in robustness and real-time.How to achieve a high level trade-off between detection accuracy and detection time in complex still or video frame images has become a major research difficulty in pedestrian detection in recent years.Therefore,in recent years,the component-based target detection algorithm has been used for pedestrian detection because of its high performance.The most effective one is the deformable part model algorithm.Deformable component model algorithm that is combines modern image features with machine learning.The main steps of pedestrian detection in this algorithm are as follows: extraction of pedestrian features,establishment of semantic and structural models of deformable components,training of deformable component models,and pedestrian location and detection on final still images.Because the feature based on gradient direction histogram is more robust when the background is complex,the illumination is stronger and the target has deformation,so the feature extraction based on the deformable component model algorithm is introduced.Gradient histogram features are generally used to describe pedestrian features.Extracted features through support vector Finally,a deformable component model is built to match the pedestrian in the target image,and then the traveller is detected and located.The algorithm of this article is based on the deformable part model.The algorithm has high accuracy,but the detection time is too long toimprove.Firstly,the convolutional neural network model is used to replace the gradient histogram feature in the feature extraction part,which solves the problem of time-consuming due to excessive computation of gradient histogram feature.Secondly,in the modeling section of deformable components,five parts are added to 8 parts,which solves the problem of reducing the accuracy rate in order to improve the detection speed.Then,in the training part of deformable component model,the hidden variable support vector machine with potential value is used to train the model,which is more accurate than support vector machine.Finally,The concatenated detection algorithm is used to build a simplified model before the pedestrian detection part,and combined with the branch and bound algorithm,the region of interest is extracted,which solves the time-consuming problem of pedestrian location.The algorithm is tested on INRIA data set.The results show that the proposed algorithm has a good detection effect on the accuracy and speed of detection.