There is a broad range of applications of visual object tracking that motivate the interests of researchers worldwide. These include video surveillance to know the suspicious activity, sport video analysis to extract highlights, traffic monitoring to analyse traffic flow and human computer interface to assist visually challenged people. In general, the processing framework of object tracking in dynamic scenes includes the following stages: segmentation and modelling of interesting moving object, predicting possible location of candidate object in each frame, localization of object in each frame, generally through a similarity measure in feature space. However, tracking an object in a complex environment is a challenging task. This survey discusses some of the core concepts used in object tracking and present a comprehensive survey of efforts in the past to address this problem. We have also explored wavelet domain and found that it has great potential in object tracking as it provides a rich and robust representation of an object. Povzetek: Podan je pregled metod vizualnega sledenja objektov .

/pdf/the-state-of-the-art-in-visual-object-tracking-eld8zbt5qi.pdf

The State-of-the-Art in Visual Object Tracking

Modified particle filter-based infrared pedestrian tracking

With the development of space missions, especially manned space missions, a reliable and secure navigation system, and assured autonomous capability in case of emergencies in space, is needed. In order to compensate for the insufficiency of ground station tracking and control, a new autonomous celestial/Doppler-integrated navigation method for a spacecraft is proposed. Celestial navigation is a fully autonomous navigation method, but in some situations the navigation accuracy of this method is subject to the inaccuracies of the measuring devices. Doppler navigation can serve as a good complement to celestial navigation. Because both the state and the measurement models of a celestial/Doppler- integrated navigation system are nonlinear and non-Gaussian, the unscented particle filter (UPF) based information fusion method is proposed here to fuse the position signals from the celestial navigation and Doppler navigation subsystems, and to enhance the navigation accuracy. The performance of this new method is tested and examined using actual spacecraft-orbit data. Simulations show that the position and velocity accuracies are estimated to within 300 m and 0.5 m s−1 respectively, which demonstrate the feasibility and effectiveness of this method. Moreover, it can be used as a backup system to provide redundancy.

http://iopscience.iop.org/article/10.1088/0957-0233/19/9/095203/pdf

Spacecraft autonomous navigation using unscented particle filter-based celestial/Doppler information fusion

Hierarchical Kalman-particle filter with adaptation to motion changes for object tracking

CNN and HOG based comparison study for complete occlusion handling in human tracking

A major challenge for most tracking algorithms is how to address the changes of object appearance during tracking, incurred by large illumination, scale, pose variations and occlusions. Without any adaptability to these variations, the tracker may fail. In contrast, if adapts too fast, the appearance model is likely to absorb some improper part of the background or occluding objects. In this paper, we explore a tracking algorithm based on the robust appearance model which can account for slow or rapid changes of object appearance. Specifically, each pixel in appearance model is represented using mixture Gaussian models whose parameters are on-line learned by sequential kernel density approximation. The appearance model is then embedded into particle filter framework. In addition, an occlusion handling scheme is invoked to explicitly indicate outlier pixels and deal with occlusion events, thus avoiding the appearance model to be contaminated by undesirable outlier ‘thing’. Extensive experiments demonstrate that our appearance-based tracking algorithm can successfully track the object in the presence of dramatic appearance changes, cluttered background and even severe occlusions.

Robust appearance-guided particle filter for object tracking with occlusion analysis

Roughly, visual tracking algorithms can be divided into two main classes: deterministic tracking and stochastic tracking. Mean shift and particle filter are their typical representatives, respectively. Recently, a hybrid tracker, seamlessly integrating the respective advantages of mean shift and particle filter (MSPF) has achieved impressive success in robust tracking. The pivot of MSPF is to sample fewer particles using particle filter and then those particles are shifted to their respective local maximum of target searching space by mean shift. MSPF not only can greatly reduce the number of particles that particle filter required, but can remedy the deficiency of mean shift. Unfortunately, due to its inherent principle, MSPF is restricted to those applications with little changes of the target model. To make MSPF more flexible and robust, an adaptive target model is extended to MSPF in this paper. Experimental results show that MSPF with target model updating can robustly track the target through the whole sequences regardless of the change of target model.

Joint tracking algorithm using particle filter and mean shift with target model updating

Existing methods of improving particle filters mainly focus on two aspects: designing a good proposal distribution before sampling and allocating particles to a high posterior area after sampling. An auxiliary particle filter (APF) is one such simple algorithm belonging to the former aspect, which generates particles from an importance distribution depending on a more recent observation. Its weakness is that it requires a large number of particles. On the other hand, a kernel-based particle filter (KPF), which belongs to the latter aspect, is able to greatly reduce the number of particles required and is still able to capture good characteristics of the posterior density. However, a KPF does not take the current observation into account. To utilize their respective strengths, a new algorithm is proposed in this paper with the combination of an APF and a KPF, the APF for designing good proposal density and the KPF for exploring the dominant mode of the posterior density. Experimental results in several real-tracking scenarios demonstrate that the integrated algorithm surpasses the standard particle filter (SPF) when encountering weak dynamic models. Moreover, the proposed algorithm is also able to achieve a comparable performance with KPF whilst reducing computational cost.

Head tracking based on the integration of two different particle filters

During object tracking, the appearance of the object being tracked often suffers from complex variations due to large illumination, pose changes, and frequent occlusions. Modeling a robust appearance model for accommodating those variations becomes a crux for appearance-based tracking algorithms. We present a tracking algorithm based on the hybrid appearance models that include a fixed appearance model, a fast-change appearance model, and an eigenspace-based model. The hybrid appearance models reveal three factors of appearance changes: stable, transient, and gradual. A particle filter is invoked to perform state inference and simultaneously be responsible for intermodel switching. In addition, robust statistics is utilized to deal with occlusion events. Numerous experiments on many difficult sequences demonstrate that our algorithm can effectively track the object undergoing large pose, lighting, and viewpoint changes. More importantly, our algorithm shows powerful ability of addressing severe occlusions.

Weifeng Tian

Papers

Robust appearance-guided particle filter for object tracking with occlusion analysis

Joint tracking algorithm using particle filter and mean shift with target model updating

Head tracking based on the integration of two different particle filters

Efficient hybrid appearance model for object tracking with occlusion handling