摘要
行人检测是计算机视觉中十分重要而又有挑战的研究方向。针对梯度方向直方图(HOG)特征描述子的局限性,如冗余信息多、容易造成误检和漏检等,为了进一步提高行人检测的准确率和速度,提出多特征融合的行人检测算法,利用主成分分析(PCA)对HOG进行降维再与局部二值模式(LBP)特征进行融合,使用支持向量机(SVM)进行分类。在INRIA行人数据库上进行测试,实验表明该算法提高了识别率,加快了训练和检测的速度。
Human detection is an important but challenging task in computer vision. For the limitation of HOG character descriptor, such as more redundant information ,likely to cause false detection and missed and so on. In order to improve the accuracy and speed of the human detection further, this paper proposes a human detection algorithm based on multi-features fusion, using PCA to reduce the dimension of raw HOG and combination it with LBP feature, using SVM algorithm for feature learning. The experimental results on INRIA data show that this algorithm training and detection speed. increases the recognition rate of human detection, the
出处
《信息技术》
2015年第2期101-105,共5页
Information Technology
