No abstract

Vision based fire detection is potentially a useful technique.With the increase in the number of surveillance cameras being installed, a vision based fire detection capability can be incorporated in existing surveillance systems at relatively low additional cost. Vision based fire detection offers advantages over the traditional methods. It will thus complement the existing devices. In this paper, we present spectral, spatial and temporal models of fire regions in visual image sequences. The spectral model is represented in terms of the color probability density of fire pixels. The spatial model captures the spatial structure within a fire region. The shape of a fire region is represented in terms of the spatial frequency content of the region contour using its Fourier coefficients. The temporal changes in these coefficients are used as the temporal signatures of the fire region. Specifically, an autoregressive model of the Fourier coefficient series is used. Experiments with a large number of scenes show that our method is capable of detecting fire reliably.

Most existing video retrieval systems use low-level visual features such as color histogram, shape, texture, or motion. In this paper, we explore the use of higher-level motion representation for video retrieval of dynamic objects. We use three motion representations, which together can retrieve a large variety of motion patterns. Our approach works on top of a tracking unit and assumes that each dynamic object has been tracked and circumscribed in a minimal bounding box in each video frame. We represent the motion attributes of each object in terms of changes in the image context of its circumscribing box. The changes are described via motion templates [4], self-similarity plots [3], and image dynamics [9]. Initially, defined criteria of the retrieval process are interactively refined using relevance feedback from the user. Experimental results demonstrate the use of the proposed motion models in retrieving objects undergoing complex motion.

We present a dynamic inference algorithm in a globally parameterized nonlinear manifold and demonstrate it on the problem of visual tracking. An appearance manifold is usually nonlinear, embedded in a high dimensional space, and can be approximated by a mixture of locally linear models. Existing methods for nonlinear dimensionality reduction, which map an appearance manifold to a single low dimensional coordinate system, preserve only spatial relationships among manifold points and render low dimensional embeddings rather than mapping functions. In this paper, we parameterize the mixture of linear appearance subspaces of an object in a global coordinate system, and apply it to visual tracking using a Rao-Blackwellized particle filter. Experimental results demonstrate that the proposed approach performs well on object tracking problem in scenes with significant clutter and temporary occlusions which pose difficulties for other methods.