محمدرضا محزون

Estimate the position of moving objects tracking is an important and Many algorithms have been proposed. In this paper, a method to estimate the position of moving objects by solving the Bayesian equations of nonlinear systems with non-Gaussian distributed algorithm based on particle filter are offered. In this way will build the first target model of weighted histogram, Then applying random noise in the location of the first frame image, predicted the candidate particles in the next step and build a histogram weighted by the candidate particles and particles Start by Bhattacharya distance weighting on the similarity between the target model and candidate model particles and estimated the target position in the next frame by the resampling algorithm in the particle filter, Finally an adaptive target model update is performed, if necessary, based on the best model for particle similar to the target.

In this paper, based on resampling particle filter algorithm, a new method to track moving object is proposed. Determining the number of levels needed for resampling in particle filter algorithm, plays an important role in the time duration of each video frame processing. By estimating the Gaussian kernel density the weighted histogram of the target model is obtained and by considering random noise variance at target place the position of candidate particles for the next frame will be predicted. In the current work, the candidate particles are weighted using Bhattacharyya distance, while the number of resampling levels is determined in accordance with the particle weights, adaptively. The radius of the kernel will be matched on moving target variations by using edge detection. Comparison of the result of proposed algorithm with the result of fixed level resampling particle filter algorithm, shows the increasing of moving target detection accuracy up to 88% without excess changes in the particles distribution variances. Moreover, the average process time decreases to 22ms.

Landmine are considered as those military weapons that when used during a war, their mission is not accomplished after the war and will always be a threatening element. Thus detection and neutralization of landmine before it causes an insident is one of the most evident examples of passive defense. On the other hand, since common methods of landmine detection face many risks, therefore selection of a method that has the lowest risk, is one of the main considerations of passive defense. Passive Infrared thermography technique which can detect buried landmines without direct contact with the soil, is one of the new and safe methods of landmine detection.In the present paper, to detect landmines, we presented a method which contains different stages of contrast stretching, image smoothing, noise reduction, image segmentation, extraction of feature image from a sequence of images and object identification. In this regard some of the image processing techniques such as linear stretching, histogram equalization, morphological filters, watershed segmentation method based on markers and Karhunen-Leove transformation have been used.

In this paper, we perform adaptive radius of the kernel by an edge detection method with tracking algorithm based on kernel density and provides a robust tracking algorithm, in combination with the resampling particle filter algorithm. In the first frame, by suitable kernel density estimation, is obtained the weighted histogram of the target model and by adding random noise variance at this place, are predicted the position of candidate particles in the next step. The weighted histogram of this candidate particles are compared with the same density kernel by the target model and are weighted the candidate particles by Bhattacharyya distance. The resampling algorithm, estimates the target position in the next frame. Finally, radius of the kernel is consistent with the target model changes. If needed, the target model is updated according to the best model of particle similar to the target model, adaptively.