Design And Implementation Of Removed Object Detection Base On Video-Sequence

With the continuous advancement of scientific and technological progress, Video Streamis the most used vector, which has played an increasingly important role in people’s daily life.Using advanced intelligent video processing technology allows a computer to help usunderstand, analyze the content of the video sequence contains and meet the people’s demandfor some specific monitoring too. Therefore, the monitoring and analysis software based onvideo sequences intelligent plays an important role in today’s scientific research and practicalapplication.The removed object detection based on video sequences use the concept of artificialintelligence to replace manual monitoring, which can achieve real-time monitoring inmulti-channel video. When a target is removed, the alarm can send the information to alertrelated people’s attention and avoid unnecessary loss of property in the first time. Therefore,removed object detection system based on video sequences has a certain theoretical value andgreat practical significance.In this paper, we use background modeling and image histogram features to achievehigh-precision detection of removed object. The main works include:First, we study the background model and foreground extraction method. In removedobject detection algorithm, the background model is the most important part. Its accuracydirectly affects whether the removed object system can be achieved. This paper describes twotypes of prospects and experimental extraction method: frame difference method andbackground subtraction method. After the analysis of the experimental results, this articledetermine integrate the first frame difference method and Gaussian background model tobuild the final background modeling method-dual background modeling method. The firstframe difference method is responsible for detecting illumination changes and settingcomfortable thresholds to control the update rate of the mixture Gaussian background model.So the introduced dual background model can quickly adapt to future changes in illumination.Second, we study the image preprocessing method based on median filter. There is a certain image acquisition noise, so we simple study a method to remove the image noise. Byanalysis, noise mostly Salt noise and Gaussian noise, so we can use median filtering methodfor noise reduction.Third, we study histogram-based method for judging the occlusion. In video surveillance,the target may be occluded intentionally or unintentionally by other targets. In this paper, byanalyzing the situation of occlusion, we use the histogram information to determineocclusions. The initial histogram feature model of monitored target can be built as a referencemodel. We also calculate the histogram feature in the current frame, then we use theBhattacharyya distance of the two models and the corresponding threshold control to achieveocclusion detection.Fourth, this paper studies the removed object detection rules. Since the detection rate ofremoved object based on foreground is not reliable, we use dual background model andhistogram information to detect removed object. Only the position of the target’s foregroundpixels exceeds a certain range and the degree of Bhattacharyya distance are tremendousexceed the threshold, the target is be judged removed. By the above method, removed targetdetection can achieve higher detection accuracy.Finally, we realize a simple removed object detection system based on the above chapters.After the relevant theoretical research and experimental results, we implements a simplesystem for detecting removed object. Through extensive testing, we verify the correctness ofthe methods and algorithms.