论文部分内容阅读
研究了利用梯度方向直方图(HOG)特征实现物体分类的方法,并且将该特征结合深度图像分割和支持向量机(SVM)分类器,实现了一个物体分类系统.该系统基于新型传感器Kinect,可以提供实时的彩色图和高精度的深度图.利用其深度图做图像分割并且还原物体的三维信息,提出了依据物体距离自适应放缩分割出的物体区域窗口尺寸的方法,解决了检测中的尺度问题.实验证明:该系统具有很高的准确度,并且在一定距离范围内具有较强的鲁棒性.
A method of object classification based on gradient histogram (HOG) is studied, and an object classification system is realized by combining this feature with depth image segmentation and Support Vector Machine (SVM) classifier.Based on the new Kinect sensor, Time color map and high-precision depth map.Using its depth map to do image segmentation and to restore the three-dimensional information of the object, a method of adaptively scaling and dividing the window size of the object region according to the object distance is proposed, Scale problem.The experiment proves that the system has high accuracy and strong robustness within a certain range.