CAJ | 학술논문

针对传统人体跟踪方法存在的不足，提出一种深度 Histograms of Oriented Gradients（HOG）和 Locally Linear Embedding （LLE）相结合的跟踪方法。首先依据图像的颜色和深度信息，结合改进的 HOG 表达式提取人体的特征向量；再利用流行学习 LLE算法对特征向量进行降维，采用欧氏距离判别法找出每帧图像人体所在区域，并对人体区域加以标记；最后，进行人体的实时跟踪。实验结果表明：降维后的人体特征数据更有助于实现跟踪，所提出的方法应用到视频图像人体跟踪中可以有效地跟踪人体，简化了数据复杂度，明显提高人体跟踪准确率。
침대전통인체근종방법존재적불족，제출일충심도 Histograms of Oriented Gradients（HOG）화 Locally Linear Embedding （LLE）상결합적근종방법。수선의거도상적안색화심도신식，결합개진적 HOG 표체식제취인체적특정향량；재이용류행학습 LLE산법대특정향량진행강유，채용구씨거리판별법조출매정도상인체소재구역，병대인체구역가이표기；최후，진행인체적실시근종。실험결과표명：강유후적인체특정수거경유조우실현근종，소제출적방법응용도시빈도상인체근종중가이유효지근종인체，간화료수거복잡도，명현제고인체근종준학솔。
Aiming at the deficiencies of traditional human body tracking method,we propose a tracking method which combines the deep histograms of oriented gradients (HOG)with locally linear embedding (LLE).First,it extracts the eigenvector of human body according to the colour and depth information of the image and in conjunction with the improved HOG expression;Then it uses the popular learning LLE algorithm to reduce the dimensionality of eigenvectors,and use Euclidean distance discriminant to find out the regions where each image frame of human body is in,and marks these regions.Finally,it makes timely human body tracking.Experimental results show that the human body feature data with dimensionality reduced is more conducive to tracking realisation,and applying the proposed method to tracking the human body in video image will be more effective,which simplifies the data complexity and significantly improves the accuracy of human body tracking.