Object Matching in Distributed Video Surveillance Systems by LDA-Based Appearance Descriptors

Abstract

Abstract. Establishing correspondences among object instances is still challenging in multi-camera surveillance systems, especially when the cameras' fields of view are non-overlapping. Spatiotemporal constraints can help in solving the correspondence problem but still leave a wide margin of uncertainty. One way to reduce this uncertainty is to use appearance information about the moving objects in the site. In this paper we present the preliminary results of a new method that can capture salient appearance characteristics at each camera node in the network. A Latent Dirichlet Allocation (LDA) model is created and maintained at each node in the camera network. Each object is encoded in terms of the LDA bag-of-words model for appearance. The encoded appearance is then used to establish probable matching across cameras. Preliminary experiments are conducted on a dataset of 20 individuals and comparison against Madden's I-MCHR is reported.

Identifying multiple objects from their appearance in inaccurate detections

“…These methods either quantize the track position and motion into visual words [14,29,34], or learn continuous distributions in the spatial domain [17]. LDA has also been used to compare person appearances in a multi-camera setup [23] from externally provided trajectories, and establish probable matches among non-overlapping viewpoints. In [19] LDA is adapted to discover behavior patterns in a multi-camera CCTV setup from quantized optical flow.…”

“…Let us now look at topic models, which originate from the unsupervised analysis for text documents, and have been successfully applied to computer vision tasks by defining these in terms of 'visual document' and a codebook of 'visual words' [26,25,23,29,19,14,17]. Latent Dirichlet Allocation (LDA) [8] is a topic model that represents a document as an unordered bag-ofwords, i.e.…”

Appearance-based person reidentification in camera networks: problem overview and current approaches

“…Gray and Tao (2008) use AdaBoost to learn a strong classifier made of week learners that are functions of image position and intensity. Lo Presti et al (2009) estimates and maintains in each node of a distributed camera network a Latent Dirichlet allocation model based on the bag-of-features representation. In (Truong Cong et al 2009 a concatenation of features is used in combination with an SVM and manifold learning to perform reidentification in a lower dimensional space to handle multiple shots.…”

An Study on Re-identification in RGB-D Imagery

“…Among the appearance cues used for this problem, interest points, structural information and color have deserved researchers attention so far [4][5][6]. Proving that 2D visual information extracted from RGB images is a valid data source to solve, at least partially, the problem.…”