Hadi Firouzi

Real-time monocular vision-based object tracking with object distance and motion estimation

This paper presents a real-time vision-based object tracking system consisting of a camera on a 2-DOF manipulator which, for example, can be a PT camera. The main novelties of the proposed tracking system include the ability to i) reduce the image processing load by relying on the object position as the only feature of the images acquired from a camera, and ii) estimate the distance and motion of the object without the need for an active rangefinder. The object tracking system is capable of controlling a manipulator using a feedback system based on the object position. The control rule of the feedback system is to minimize the distance between the object position in the camera image and the center point of the image. The proposed method can be readily adopted in dynamic environments, and achieve better tracking accuracies and efficiencies than the traditional methods. The formulation of the distance and motion estimation is presented in the paper. The efficiency and robustness of the proposed method in dealing with noise is verified by simulation using a PT camera.

Adaptive on-line similarity measure for direct visual tracking

This paper presents an on-line adaptive metric to estimate the similarity between the target representation model and new image received at every time instant. The similarity measure, also known as observation likelihood, plays a crucial role in the accuracy and robustness of visual tracking. In this work, an L2-norm is adaptively weighted at every matching step to calculate the similarity between the target model and image descriptors. A histogram-based classifier is learned on-line to categorize the matching errors into three classes namely i) image noise, ii) significant appearance changes, and iii) outliers. A robust weight is assigned to each matching error based on the class label. Therefore, the proposed similarity measure is able to reject outliers and adapt to the target model by discriminating the appearance changes from the undesired outliers. The experimental results show the superiority of the proposed method with respect to accuracy and robustness in the presence of severe and long-term occlusion and image noise in comparison with commonly used robust regressors.

Robust PCA-Based Visual Tracking by Adaptively Maximizing the Matching Residual Likelihood

A new similarity measure called matching residual likelihood (MRL) is presented for the task of visual tracking. MRL estimates the likelihood of the matching residual between the object representation model and the new candidate image based on previous matching errors. A posterior probability called a posterior matching residual probability is modeled based on the matching residual likelihood, the object motion model between sequential states, and the prior probability to estimate the density distribution of the object location. At every frame, an on-line algorithm is used to learn a low-dimensional PCA model from the object image. Then the object is located by maximizing the posterior matching residual probability distribution of the object state based on a robust factored sampling algorithm. The proposed method cab readily update the similarity measure to handle significant appearance changes while it is still robust to outliers and occlusion. In our experiments, the proposed tracker is applied on several challenging image sequences and the result is compared with other state-of-the-art methods and the ground truth data. The comparison results show the robustness and accuracy of our tracker in existence of large object motion and appearance variation, occlusion, outliers, and illumination changes.

Adaptive Non-rigid Object Tracking by Fusing Visual and Motional Descriptors

This paper presents a framework to track non-rigid objects adaptively by fusion of visual and motional feature descriptors. The proposed technique can automatically detect an object from different points of view as soon as the object starts moving. Moreover an object model is created and gradually updated using both new and previous features. As a result, the proposed technique is able to track a non-rigid object even if the object is rotating or distorting. The efficacy of the proposed method is verified using the experimental results obtained from a grayscale camera.

Efficient and robust multi-template tracking using multi-start interactive hybrid search

This paper presents an efficient, accurate, and robust template-based visual tracker. In this method, the target is represented by two heterogeneous and adaptive Gaussian-based templates which can model both short- and long-term changes in the target appearance. The proposed localization algorithm features an interactive multi-start optimization process that takes into account generic transformations using a combination of sampling- and gradient-based techniques in a unified probabilistic framework. Both the short- and long-term templates are used to find the best location of the target, simultaneously. This approach further increased both the efficiency and accuracy of the proposed tracker. The contributions of the proposed tracking method include: (1) Flexible multi-model target representation which in general can accurately and robustly handle challenging situations such as significant appearance and shape changes, (2) Robust template updating algorithm where a combination of tracking time step, a forgetting factor, and an uncertainty margin are used to update the mean and variance of the Gaussian functions, and (3) Efficient and interactive multi-start optimization which can improve the accuracy, robustness, and efficiency of the target localization by parallel searching in different time-varying templates. Several challenging and publicly available videos have been used to both demonstrate and quantify the superiority of the proposed tracking method in comparison with other state-of-the-art trackers.

Multiple object tracking via a two-way confidence-based correspondence algorithm

In this paper an efficient object correspondence algorithm is presented for tracking multiple objects in dynamic environments is proposed. It is assumed that objects can be added to the scene, removed from the environment, or occluded by other objects. The proposed algorithm benefits from two key features including a confidence measure and a two-way matching mechanism to improve the correspondence accuracy. Unlike the traditional methods which solve the correspondence problem by matching new objects and then removing the incorrect ones, our algorithm avoids establishing invalid correspondences considering a confidence measure. Also, we introduce a two-way correspondence algorithm consisting of forward matching and backward matching. As a result, the track of objects is expanded both from the head and tail using new objects and previous unmatched objects respectively. The proposed method has been applied to synthetic data, and the results show the efficiency and reliability of the method against a large number of objects.

Robust gaussian-based template tracking

In this paper a visual object tracking method is presented which is robust against changes in the object appearance, shape, and scale. This method is also able to track objects being occluded temporarily in cluttered environments. It is assumed the target object moves freely through an unpredicted pattern in a dynamic environment where the camera may not be stationary. The proposed method models the object representation by an adaptive and deformable template which consists of several Gaussian functions. A 5 degree-of-freedom transformation function is employed to map the pixels from the template reference frame to the image reference frame. Moreover, the object localization method is based on a robust probabilistic optimization algorithm which is performed at every image frame to estimate the transformation parameters. The comparisons of the results obtained by the proposed tracker and several state-of-the-art methods with the manually labeled ground truth data demonstrate higher accuracy and robustness of the proposed method in this work.

Detection and tracking of multiple similar objects based on color-pattern

In this paper an efficient and applicable approach for tracking multiple similar objects in dynamic environments is proposed. Objects are detected based on a specific color pattern i.e. color label. It is assumed that the number of objects is not fixed and they can be occluded by other objects. Considering the detected objects, an efficient algorithm to solve the multi-frame object correspondence problem is presented. The proposed algorithm is divided into two steps; at the first step, previous mismatched correspondences are corrected using the new information (i.e. new detected objects in new image frame), then all tail objects (i.e. objects which are located at the end of a track) are tried to be matched with unmatched objects (either a new object or a previously mismatched object). Apart from the correspondence algorithm, a probabilistic gain function is used to specify the matching weight between objects in consecutive frames. This gain function benefits Student T distribution function for comparing different object feature vectors. The result of the algorithm on real data shows the efficiency and reliability of the proposed method.