项目名称: 视频目标语义交互模型的人行为检测的关键技术研究
项目编号: No.61272439
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 蒋兴浩
作者单位: 上海交通大学
项目金额: 82万元
中文摘要: 对视频中人的行为检测的研究,属于信息安全和信息智能分析理解领域的应用基础研究。对复杂视频环境下人行为的正确识别和理解,对预测和事后分析某些特殊事件,具有非常重要的意义。本课题围绕新的视频特征获取方法和新的目标语义交互模型,展开对视频中人的交互行为检测算法研究。在视频新特征获取的方法中,提出基于改进HOG特征的人形检测算法、提出高速HOG的优化算法、基于运动树先验概率模型的人体关节分析方法等关键技术。还提出基于目标区域概率模型的人体与物体语义交互识别的模型,可更加精确地检测人的行为。课题提出的基于目标语义交互模型的人行为检测的关键技术研究,有效融合了人关节、肢体的特征和目标语义交互模型的各自优势,提高对视频中人的各种行为检测的识别准确性和适用范围。研究成果对城市公共安全监控、海量监控视频关联性分析、特定场景下的个体行为分析及预测等应用,具有显著的社会效益。
中文关键词: 人体行为检测;交互模型;模式识别;异常检测;机器学习
英文摘要: Human action detection technology is the application of basic research in the field of multimedia information processing technology and content security technology. It is an important potential applications to detect human action accurately and efficiently for understanding the human action and events and analyzing the video content retrieval precisely. The topic expands the video human action detection algorithm based on new video features extracting methods and new target semantic interaction models. In our new methods of video features extracting, we proposed a humanoid detection algorithm based on improved HOG. A high-speed HOG optimization algorithm is proposed based on the key technology of the prior model of human movement tree joint analysis methods. Besides the semantic interaction of human body and object recognition model is trained based on the probabilistic model of the target area, thus improving the accuracy rate of human action detection. The proposed human action detection methods based on the interaction model of the target semantic integrates the respective strengths of human joints, limb characteristics and the objective semantic interaction model effectively to improve the recognition accuracy of the video human action detection and expand the range of applications. It is more efficient to u
英文关键词: Human Behavior Detection;Interaction Model;Pattern Recognition;Abnormal Detection;Machine Learning