Improved Multi-feature Fusion Algorithm for Target Detection Based on Haar-like and LBP

Abstract

Aiming at the problems of detecting targets with Haar-like,such as excessive feature values,expensive computational cost,inability to describe the target texture features,and low recognition rate etc.,this paper proposes an adaptive threshold IHL (Improved Haar-like LBP) feature extraction algorithm based on the information of the origin of the sliding window.More speci-fically,the algorithm first constructs the IHL feature coding method fusing Haar-like features and LBP features.Then,for computing Haar-like local binary features,a Gaussian matrix is used to obtain an adaptive threshold that conforms to the pixel distribution law,and the pixel information of the central point is introduced to ensure the rationality of the extracted feature value.Finally,the cascade classifier trained by AdaBoost is built and experiments are conducted on the KITTI vehicle data set and the INRIA Person pedestrian data set.The proposed algorithm,with a recognition rate of more than 94%,is proven effective with recognizing 1 102 pedestrian targets in 65 s and 1 852 vehicle targets in 114.3 s,which can significantly speed up target recognition time and obviously improve the detection accuracy compared with state-of-the-art methods.

Keywords

• target detection|multi-feature fusion|feature extraction|haar-like features|ihl characteristics