| Video monitoring system which has a broad development prospect is an importantresearch content in the fields of computer vision, pattern recognition and artificialintelligence where moving target detection and background modeling are the mostimportant research direction. Moving target detection which is the key link of thesubsequent moving target recognition and tracking is to extract the moving changed areafrom the background of the video sequences for the subsequent processing. And it is still aresearch hot that how to detect the moving target efficiently and accurately in the field oftarget recognition and tracking.Moving target detection algorithms are various. In this paper, firstly, some commonmoving target detection algorithms are simply introduced including the backgroundsubtraction division method, frame difference method and the optical flow method. Andbackground subtraction division method which has the characteristic of easy realization,good real-time and high accuracy is the most popular. Secondly, a nonparametric densityestimation background difference method is proposed in this paper. Next, the advantagesand disadvantages of each algorithm are summarized after analyzing the experiments.Finally, the kernel density estimation method of the nonparametric density estimationmethod is selected as the research key.As one of the most common used background modeling method, the nonparametrickernel density estimation is widely used in the background modeling aspects for its fastadaptive to the changing dynamic scenes. However, it is difficult for the method to satisfythe real time requirement of the moving target detection for its high complexity. After thefurther study of the nonparametric kernel density estimation model, we find that the modelestimates the probability model of pixels using the history sampling values of each pixel.Remarkably, the model may have high complexity and error estimation because the originalhistory sampling values contain lots of redundant information and noise information. Inview of these shortcomings, some improved ideas of the nonparametric kernel densityestimation are put forward in this paper. We can build background model and update thebackground by extracting feature frames because the kernel density estimation need notpre-assume the density distribution function of the background model. The construction of new samples not only reduces the calculated amount of the kernel density estimation andimproves the speed of the detection, but also reduces the false alarm rate and the errordetection rate of target detection.In conclusion, some solutions of the problems of using the kernel density estimationmethod to detect moving targets are put forward in this paper after further studying themethod. The experiments show that the proposed algorithm is effective and reliable. |
