摘要
为了提高在复杂环境下行人检测的准确率,提出一种基于改进可变形部件(DPM)模型的行人检测方法。特征提取时,将图像单元梯度方向离散到9个区间中,采用解析降维法代替主成分分析法对归一化的特征进行降维;将行人的身体部分分为多个模块,提取出行人以及各个模块的特征,然后分别针对行人以及模块进行分类器训练得到根模型与模块模型;最后采用Cascade方法代替传统的动态规划法对行人检测。在Matlab环境下进行仿真,结果表明该方法与传统的基于HOG特征的检测方法相比,识别率得到了明显提高,而与原始DPM方法相比,在不降低识别率的同时,能够在一定程度上提高其检测速度。
In order to improve the accuracy of pedestrian detection in complex environment, the paper proposed a pedestrian detection method based on improved DPM model. When the feature is extracted, to reduce the dimension of the normalized feature, the reduced dimension method is used instead of the principal component analysis, and the gradient direction of the image unit is discretized into 9 intervals. The pedestrian is divided into several modules based on body parts, and the features of pedestrian and each modules are extracted. Then, the root model and module model are obtained by classifier training for pedestrians and modules respectively. Finally, the Cascade method is used instead of the traditional dynamic programming method for pedestrian detection. The results show that with the method of simulation in Matlab environment, the recognition rate is significantly improved compared with the traditional detection method based on HOG feature, and compared with the original DPM method, it improves the detection speed to a certain extent without reducing the recognition rate.
作者
张亚须
龙晖
云利军
Zhang Yaxu;Long Hui;Yun Lijun(School of Information,Yunnan Normal University,Kunming 650500,China;Information Center of General Office of Yunnan Provincial Committee,Kunming 650021,China;Yunnan Key Laboratory of Optoelectronic Information Technology,Kunming 650500,China)
出处
《大理大学学报》
CAS
2018年第6期13-18,共6页
Journal of Dali University
