Multiple video object tracking using variational inference

Abstract

In this article a Bayesian filter approximation is proposed for simultaneous multiple target detection and tracking and then applied for object detection on video from moving camera. The inference uses the evidence lower bound optimisation for Gaussian mixtures. The proposed filter is capable of real time data processing and may be used as a basis for data fusion.The method we propose was tested on the video with dynamic background,where the velocity with respect to the background is used to discriminate the objects. The framework does not depend on the feature space, that means that different feature spaces can be unrestrictedly used while preserving the structure of the filter.

Dense Compounded Features and Data Association Based Multiple People Tracking

“…Many studies have been carried on multiple object tracking. A Kalman filter has been used by [11], [12], [13] to introduce MOT system. The studies were based on RANSAC, 1 st Mahmoud Al-Faris 2 nd John Chiverton background and shape, and mixture of Gaussian matching algorithms respectively.…”

Intelligent Video Surveillance

“…two novel methods for object detection and tracking have been developed using the newly proposed Bayesian filtering technique [15], [16], the object detection and tracking techniques were applied and evaluated on the data sets and applied to thermal and optical video data; -the evolving classifier AutoClass has been developed and evaluated on the image data [17]; -the TEDA framework has been exploited for building up the new classifiers and clustering algorithms [18], [19], [20], [21]; -the Chan-Vese image segmentation algorithm has been improved which lead in significant increase of the algorithm speed, by fitting a Chan-Vese functional to a Boolean programming problem. Applications of the algorithm to medical image analysis were also investigated [22]; -a book chapter has been published jointly with two other co-authors, Denis Kolev and Mikhail Suvorov, describing SVM-based methods, in [23].…”

“…The method is presented in two versions: featuring Laplacian and Variational approximations on the update step of the Bayesian filter. The first method was presented at the Fusion 2015 conference [15], while the second one was presented on Sensor Data Fusion [16].…”