Md. Zahidul Islam

Real Time Moving Object Tracking by Particle Filter

Robust and real time moving object tracking is a tricky job in computer vision problems. Particle filtering has been proven very successful for non-Gaussian and non-linear estimation problems. In this paper, we first try to develop a color based particle filter. In this approach, the object tracking system relies on the deterministic search of window, whose color content matches a reference histogram model. A simple HSV histogram-based color model is used to develop our observation system. Secondly and finally, we describe a new approach for moving object tracking with particle filter by shape information. The shape similarity between a template and estimated regions in the video scene is measured by their normalized cross-correlation of distance transformed image. Our observation system of particle filter is based on shape from distance transformed edge features. Template is created instantly by selecting any object from the video scene by a rectangle. Experimental results have been presented to show the effectiveness of our proposed system.

A gesture recognition interface with upper body model-based pose tracking

This paper presents a gesture recognition interface with the observed pose sequence determined by our upper body model-based pose tracking. For last decade many researchers have focused on how well tracks human poses based on predefined pose model. Then we move this discussion to the gesture recognition by pose tracking. Our system consists of two parts: pose tracking and gesture recognition. In the first part, Particle filtering is used for tracking the upper body pose with the key pose library where we try to find key pose for the proposal distribution. The particles generated from the proposal distribution with random noise, could cover the pose space which is not covered by the key pose library. In second part, the observed pose is labeled with a pose number among all key poses by comparing between key poses. HMM is used to determine the probabilities of the gesture states from the observed pose sequence. HMM parameters like transition and emission matrix are trained by the analysis on the gesture database. The experimental results shows how well gesture recognition works based on our system.

MRF-based Particle Filters for Multi-touch Tracking and Gesture Likelihoods

Multi-touch tracking algorithm requires maintaining separate identities for multi-touch points, however, it fails when independent particle filter for each object is kidnapped by neighboring targets. This is called the hijacking problem. The motion model using Markov random field (MRF) has been proposed for avoiding this problem by lowering the weight of particles which are close to neighboring touch points. This paper improves the MRF-based particle filters for multi-touch tracking by optimizing the distance of neighboring touch points to reduce hijacking problem. In experiments, the optimum distance is around 80 pixels, which exhibits highly robust and optimized multi-touch tracking. Additionally we discuss about the simultaneous estimation of gesture likelihoods with MRF potentials from the tracking results.

Pictorial structures-based upper body tracking and gesture recognition

Tracking the articulated human body has been a difficult research because body poses change so dynamic and vary in visual appearance. Pictorial Structures (PS) with dynamic programming (particle filtering) has been widely used for tracking human body, which is highly articulated and moves dynamically. In this paper, we use PS and a particle filter for upper body tracking. However, a Markov-process-based dynamic motion model for particle filtering cannot adequately predict the particles. We propose a key-pose-based proposal distribution that uses similarities between the input silhouette image and the key poses to effectively predict the particles. We select relatively few example poses from the pose space as key poses, train for embedded features, and formulate the proposal distribution with key pose similarities and a Markov-process-based dynamic model. We experimentally evaluate our proposal method and an observation model and test gesture recognition for human-robot interaction.

Multi-part histogram based visual tracking with maximum of posteriori

This paper presents a new multi-part histogram (MPH) based algorithm for non-rigid object tracking with Particle filtering state estimation approach. The reference and target object are represented by some sub-region with integral image technique. Each region has its own histogram and we calculate the weight of each particle based on its region position of target object. The most weighted particle settles on the centre position of the bounded target and gradually decreases the weight of particle vertically and horizontally from the centre position. For smooth tracking Maximum a Posteriori (MAP) is introduced for better observation likelihood calculations. Experiments of proposed tracker show that our system is robust against false target tracking, severe occlusion, and rotation, scaling without extra computational cost.

Articulated hand tracking using key poses driven particle filtering

Tracking an articulated hand is very difficult problem due to the high dimensionality of the hand joint movements. We propose a system to track the articulated hand using our key pose driven particle filtering. The articulated hand is modeled as a cardboard model which has 24 DOF. Using motion constraints between the finger joints, the dimension of the articulated hand model is reduced to 13 DOF. The proposal distribution is based on the Gibbs sampler-based motion model and the matching probabilities of key poses onto the observation image. The motion model is based on the motion constraints between the hand joints. Each movement of joints is predicted by Gibbs sampler which is modeled as our motion model. We show the experimental results of tracking the articulated hand by Key poses driven particle filtering.

New Integrated Framework for Video Based Moving Object Tracking

In this paper, we depict a novel approach to improve the moving object tracking system with particle filter using shape similarity and color histogram matching by a new integrated framework. The shape similarity between a template and estimated regions in the video sequences can be measured by their normalized cross-correlation of distance transformation image map. Observation model of the particle filter is based on shape from distance transformed edge features with concurrent effect of color information. The target object to be tracked forms the reference color window and its histogram are calculated, which is used to compute the histogram distance while performing a deterministic search for matching window. For both shape and color matching reference template window is created instantly by selecting any object in a video scene and updated in every frame. Experimental results have been offered to show the effectiveness of the proposed method.

Adaptive template based object tracking with particle filter

In this paper, we describe a new approach to improve the video based object tracking system with particle filter using shape similarity. It deals with single object tracking whose dynamics age highly non-linear. The shape similarity between a template and estimated regions in the video sequences can be measured by their normalized cross-correlation of distance transformation. Here within this present job, observation model of the particle filter is based on shape from distance transformed edge features. Template is created instantly by selecting any object in a video scene and updated in every frame. Experimental results have been offered to show the effectiveness of the proposed method.

Modifed Re-sampling Based Particle Filter for Visual Tracking with MPH

In this paper we propose a particle filter based strategic approach to enhance the performance of visual tracking system with a new re-sampling algorithm. In any particle filter based application especially in visual tracking system, re-sampling is a vital process in the implementation of particle filtering. Usually it is a linear function of particle weight calculation to know the number of particle copies. But, in visual tracking problem it makes the system computationally expensive to make real time performance. We can take more effective particle which has high weight by changing re-sampling as non-linear function. For smooth tracking Maximum a Posteriori (MAP) is introduced for better observation likelihood calculations. Experiments of proposed tracker show that our system is robust against false target tracking, severe occlusion, and rotation, scaling without extra computational cost. We evaluate our system with MPH measurement techniques, and demonstrate that, our proposed new re-sampling algorithm works well.

Attenuated sequential importance resampling (A-SIR) algorithm for object tracking

This paper presents a newly developed attenuating resampling algorithm for particle filtering that can be applied to object tracking. In any filtering algorithm adopting concept of particles, especially in visual tracking, re-sampling is a vital process that determines the algorithms performance and accuracy in the implementation step.It is usually a linear function of the weight of the particles, which decide the number of particles copied. If we use many particles to prevent sample impoverishment, however, the system becomes computationally too expensive. For better real-time performance with high accuracy, we introduce a steep Attenuated Sequential Importance Re-sample (A-SIR) algorithm that can require fewer highly weighted particles by introducing a nonlinear function into the resampling method. Using our proposed algorithm, we have obtained very impressive results for visual tracking with only a few particles instead of many. Dynamic parameter setting increases the steepness of resampling and reduces computational time without degrading performance. Since resampling is not dependent on any particular application, the A-SIR analysis is appropriate for any type of particle filtering algorithm that adopts a resampling procedure. We show that the A-SIR algorithm can improve the performance of a complex visual tracking algorithm using only a few particles compared with a traditional SIR-based particle filter.