【作者】 肖本贤； 陆诚； 陈昊； 余炎峰； 陈荣保；

【Author】 Xiao Benxian,Lu Cheng,Chen Hao,Yu Yanfeng,Chen Rongbao Institute of Industrial Automation,Hefei University of Technology,Hefei 230009,P.R.China

【机构】 合肥工业大学自动化研究所；

【摘要】 针对视频监视的复杂背景,提出了一种基于帧间差分法和不变矩特征的运动目标检测与识别方法。在运动目标检测算法中,首先对定义的象素区域进行数据分析,然后通过帧间的数据差异产生运动信号,捕捉到场景中的运动目标。在目标识别方法中,首先提取运动目标区域的不变矩特征,并对其矢量进行标准化,然后利用遗传小波神经网络作为模式识别器,实现运动目标的自动识别。为验证本文方法及网络模型的泛化能力,在Matlab7.0中对模型进行了仿真。仿真实验表明,这是一种快速有效的运动目标检测与识别方法。更多还原

【Abstract】 In order to the complicated background of video monitoring system,a method of moving object detection and recognition was proposed based-on the frame difference algorithm and moment invariant features.In this moving object detection algorithm,data analysis was done for defined pixel region firstly,then the moving signal was produced by the frame data difference,and the moving object was captured from natural scene image sequences.In object recognition algorithm,moment invariant features were extracted from moving object region firstly,and vector standardization was done for these moment invariant features,then wavelet neural network with genetic algorithm was used as pattern recognition and automatic recognition was realized for moving object.In order to demonstrate above-mentioned method and the generalization ability of network model,simulation was done for this model in Matlab7.0.Simulation results show that the proposed approach is a fast and effective method for moving object detection and recognition.更多还原