Research On Moving Target Detection And Shadow Elimination Algorithms In Complex Background

With the continuous progress of the information age,people are pushing for intelligent video surveillance systems more and more urgently.The moving target detection algorithm requires fast and accurate separation of moving targets in the video sequences from complex backgrounds.Its results directly affect the function of follow-up target recognition and behavior analysis algorithms under intelligent video surveillance system,so the research of moving target detection algorithm has important application value.At present,classical moving object detection algorithms can accurately detect moving objects in simple scenes,but there are many interference factors that affect the detection results in complex environments.In order to solve the problems such as ghost,dynamic background and intermittent motion of moving target detection algorithm under practical complex scenes,this thesis designs an Adaptive Model Size Background Extractor(AMSBE).Aiming at the shadow problem prevalent under practical environment,a motion shadow elimination algorithm combining color and local five similarity pattern(LFSP)texture features is designed.This thesis includes the following main points:(1)For the case that the number of fixed-size background samples is not suitable for different pixels in the scene,this thesis proposes an adaptive background sample number based on the complexity of the pixels.In our algorithm,a large number of background samples are obtained in the dynamic background region,while the number of background samples in the static background region is relatively small.This strategy takes the time consumption and detection effect of the algorithm into account.(2)In order to solve the dynamic background problems such as leaf sloshing,this thesis improves the adaptive decision threshold and background model update rate by feedback system based on Pixel-Based Adaptive Segmenter(PBAS)algorithm.It has a high decision threshold in the dynamic background area,which solves the problem of false detection in the dynamic background area.At the same time,the moving target area obtains a low background model update rate,which avoids pollution caused by updating the moving target information into the background model.(3)Considering that random selection and first-in-first-out updating strategies cannot accurately select background samples that need to be updated in the process of background model updating,this thesis adopts the updating strategy based on the validity of background samples.The background samples with the lowest validity are selected for updating,which ensures the validity of the background model.(4)Aiming at the ghost problem caused by moving objects when the background model is initialized,this thesis improves the probability updating algorithm based on foreground counting matrix to eliminate ghost.The idea of multi-threshold is adopted.so the foreground counting matrix values of pixels located in different threshold ranges have different updating rates of background model.Two problems have been solved,one is that small single threshold selection will regard the slow moving targets as ghost which cause pollution to background model,the other is that large single threshold selection leading to long time existence of the ghost.(5)Aiming at the shadow problem generally exists in actual scenes,this thesis proposes a shadow elimination algorithm based on local Five Similarity Pattern(LFSP)texture features.And the motion shadow elimination algorithm combining the color and LFSP texture features is designed.The HSV color space and LFSP texture features are used to obtain the respective shadow detection results.The test results are obtained by combining the above two shadow detection results.The experimental results show that the AMSBE algorithm in this thesis can overcome the problems like dynamic background,ghost problem and target intermittent motion.The experimental results of eight test videos on CDnet datasets show that the comprehensive evaluation index F-measure of the AMSBE algorithm is 30% higher than that of ViBe algorithm.The multi-feature fusion motion shadow elimination algorithm can guarantee low miss detection rate and false detection rate at the same time.The comprehensive evaluation index Avg in the three CDnet dataset test videos is improved by 3.4% when compared with the HSV color space based shadow elimination algorithm.