Since the common learning model of traditional support vector machine (SVM) can't adapt to the change target posture in video,and is limited with occlusion or complex background,a new SVM neighborhood learning model is proposed.The neighborhood learning is based on the high correlation of video adjacent frame both in time and space.The training data of each testing frame is extracted from its adjacent frame to learn and update the SVM model.With the updating of the video,the SVM model is updated continually to adapt to the changes of target detection.A large number of samples were experimented in complex environment.The statistics is used to analyze the results to verify that the SVM neighborhood learning has higher accuracy and better robustness than the traditional SVM common learning.%针对传统SVM普通学习模型无法适应视频中目标姿态变化、有遮挡或复杂背景的局限性,提出一种新的SVM邻域学习模型.邻域学习是基于视频相邻帧在时间和空间上的高度相关性,每个测试帧在其相邻帧上抽取训练数据进行SVM模型的学习与更新,随着视频的更新,SVM模型将不断更新来适应目标检测的各种变化.通过大量样本在各种复杂环境下实验,采用统计学分析结果,证明SVM邻域学习比传统SVM普通学习准确率更高、鲁棒性更好.
展开▼
