Vehicle Detection Algorithm Based on Modified Gradient Oriented Histogram Feature

• Main Authors: Cing-De Su, 蘇慶德
• Other Authors: SHEU,MING-HWA
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/4329xy

碩士 === 國立雲林科技大學 === 電子工程系 === 104 === In recent years, the public safety and home security are more and more important. The surveillance system will be becoming a hot industry. Therefore, this thesis proposed a modified gradient oriented histogram feature to identify vehicle for effective traffic control. This proposed method is divided into two parts. The first part is vehicle algorithm which use positive and negative samples to be input images in the training. The principal direction and the direction histogram are used for classification characteristics. Each pixel in the oriented image is represented by an angle bin, and 8*8 pixels for a cell histogram calculated is the presented by a 6*6 cell direction histogram. According the direction histogram, the maximal number of direction is the principal direction. The modified histogram orientation gradient (MHOG) feature is obtained by overlapping two cell in the cell direction histogram. The training parameters are obtained by inputting the MHOG features to SVM. When the principal direction of input image is same with the principal direction of training image, and the decision function of SVM is 1. Then, the window image will be a vehicle image. Experimental results show that the vehicle detect algorithm to achieve 98% which is better than SVM by HOG Feature detection. And average executing velocity of our method increase 40% in computer.