Background Extraction Of Complex Scene Based On Adaptive Features

With the rise of intelligent monitoring equipment,background extraction is widely used in the fields of intelligent transportation,video surveillance,human-computer interaction and medical diagnosis.The purpose of background extraction is to extract the objects of interest in the video.Video processing using a computer can avoid the visual limitations of the human eyes.For example,in the complex scenes,human cannot accurately recognize the moving foreground target,and the real-time property required in the video analysis cannot be guaranteed.At present,background extraction has achieved good results in extracting moving targets in simple scenes,but in complex real-world environments,such as lighting changes,shadows and motion backgrounds,the extracted targets have noises and lower accuracy.Aiming the problems of background extraction in complex dynamic scenes,we proposed a background extraction model based on adaptive features for complex scenes.Firstly,the grayscale,RGB color and texture features of the video frame pixels are extracted.The extracted features are represented in the form of a histogram in time series,and the weights of the features are determined by a histogram.Secondly,the extracted features are constructed into feature vectors,and the background model is further constructed by combining the weights of features.Then,the measured pixel is adaptively matched with the background model and a double threshold comparison method is adopted.For the value between thresholds,the foreground pixels are detected in combination with the neighboring pixel points.Finally,the background model is updated in real time using the model parameter update and spatial neighborhood update strategy.The experiments show that the background extraction method of adaptive features is used to detect foreground moving targets in the changedetection.net(CDN)video library.We combined the recall and precision evaluation indicators,its comprehensive evaluation reached more than 90%.At the same time,compared with classical algorithms such as hybrid Gaussian method,we proposed algorithm has good accuracy and robustness for foreground target detection of complex dynamic background.