Research On Human Target Recognition Algorithm Based On Feature Fusion

In recent years,the development of automatic driverless cars,robotic intelligent sensing,security monitoring and other fields has made human target recognition technology a hot research topic and an important technical basis.However,in the process of identification,due to the non-rigidity of the human target and the interference of the surrounding environment,noise and occlusion of the image target cause problems in the accuracy,false detection rate and missed detection rate of the human target recognition.To this end,based on the method of feature fusion,in the process of human target recognition,the description of the target feature information is single,the occlusion target is not recognized,and the performance of the classifier is poor,so as to obtain a better recognition effect.The specific work is as follows:1.A multi-feature adaptive weighted fusion method is designed.The problem of the difference between the background and the target information is improved by the Histogram of Oriented Gradient(HOG)feature.By changing the variance of the gradient values in the cell unit,the difference between the target and the background is enhanced,and the difference is achieved.Principal Component Analysis(PCA)method for dimensionality reduction of feature information;at the same time,the original Local Binary Pattern(LBP)is improved,the sampling mode is changed,and the central pixel is compared with the surrounding pixels.The mean of the values replaces the initial threshold.Finally,the two improved features are serially combined by adaptive weighting,and the experimentally proved that the merged features have good target description ability.2.A Deformable Part Models(DPM)human body recognition algorithm based on feature fusion is designed.By analyzing the composition and response score calculation methods of DPM,the weighted component model is designed,and the problem of target overlapping occlusion is solved by assigning different weights to each component model.At the same time,there is a pair of traditional HOG features due to the use of a single HOG feature.The problem of insufficient description of target information,this paper uses the second chapter feature fusion method to replace the original single HOG feature,thus improving the effective description of human targets.3.A human body recognition algorithm based on Support Vector Machine(SVM)parameter optimization is designed.A linear combination of kernel functions is designed to replace the single kernel function in the traditional SVM classifier.For the outliers appearing in the sample data,the penalty factor is set by introducing the slack variable to eliminate.The five parameters generated by the improved SVM in the design process are rationally optimized,and the combination of particle swarm optimization and K-weight cross-validation is used to optimize the SVM classifier and feature fusion.Identification,experiments show that the optimized SVM classifier can identify human targets in complex environments.