时频降噪在图像序列事件检测中的应用

Temporal-frequency denoising application in event detection base on image sequences

针对图像序列分析中事件检测在强背景噪声的野外应用场景下经常出现'假事件'检测的问题,利用Kalman滤波器与二维DCT滤波思想,提出了基于图像序列时频降噪的事件检测方法。该方法基于Kalman滤波器对图像序列进行背景模型的时域多帧降噪,并结合变化前景区域实现背景模型的自适应重构,对单帧前景图像应用二维DCT变换实现低通降噪,最后由自适应分割方法实现事件前景的分割。通过对实际采集的野外图像序列的仿真分析表明,该方法较好地克服了'假事件'检测的问题,并更好地保持了真实事件信息,其F-measure达0.9423,具有很好的实用性与鲁棒性。

To solve the ＂false event＂ detection problem in image sequence analysis in the wild situation with strong background noises,an event detection method based on temporal-frequency denoising is proposed utilizing Kalman filter theory and 2D Discrete Cosine Transform（DCT） theory.The method utilizes Kalman filter to denoise the background model of image sequence with several frames in time domain.The background model is reconstructed adaptively based on variational foreground region.Then,the foreground image of a single frame is denoised utilizing 2D DCT transform.Finally,the event foreground is segmented by adaptive segmentation method.Based on the results of simulation analysis of image sequences collected in wild situation,the proposed method is proved to detect the event foreground and solve the ＂false event＂ detection problem effectively and practically,while its F-measure can achieve 0.9423.