Article,
OBJECT TRACKING BASED ON BAYESIAN MONTE CARLO EMPLOYING PARTICLE GAUSSIAN INFORMATION FILTER
International Journal of Advance Robotics & Expert Systems (JARES), 1 (6): 11 - 21 (February 2020)
Abstract
VISUAL tracking is one of the rapidly developing fields of computer vision. In visual field, Object tracking is a significant task in various computer vision applications like surveillance, augmented reality, humancomputer interfaces and medical imaging. Moving object detection and tracking is an essential image processing analysis in several applications for crowd monitoring. Moving object detection is utilized to enhance the image processing analysis. The object detection method is employed to detect the moving object areas with different size of objects and video progression. Then, Object tracking is vital condition for every logical video surveillance system. The existing work presented a Cognitive Control Inspired Approach (CCIA) for extended targets. Here, the two Kalman filters are used in visual tracking systems to predict the object motion depends on it size. Cognitive Perceptor unit measurement is processed from environment field to make representation of external world. Cognitive Control Unit described a set of actions and it evaluates the hidden variable by Hidden Markov Model. However, object tracking was difficult and consume more time in outdoor environments. In addition to, redundant and unwanted information were not removed with higher signal to noise ratio. The performance of object tracking accuracy was not efficiently enhanced. In order to solve the above problems, Bayesian Monte Carlo employing Particle Gaussian Information Filter scheme is proposed for object tracking to increase the accuracy for tracking the object. Bayesian approach is used for state estimation to reduce the signal to noise ratio. Initially, the Bayesian State Estimator using Monte Carlo Particle Simulation (MCPS) is estimated the set of particles (i.e. objects) with
associated weighted via posterior density. As a Bayesian estimator, particle simulation considered two main steps namely prediction and update by using particle information filter for object tracking. Then, Monte Carlo Particle State Estimator algorithm is employed for achieving the object tracking accuracy from state estimation by removing unwanted and redundant information. Finally, the Gaussian Information Filter scheme is applied with the estimated state to track the original object with noise reduction. Gaussian Information Filter scheme is described by information matrix and information vector to detect the multiple moving objects. The performance of proposed Bayesian Monte Carlo employing Particle Gaussian Information Filter scheme is analyzed against with the following metrics such as Signal-to-noise ratio, Object tracking accuracy and Mean square error with respect to number of objects.
