Learning Compact Binary Codes for Visual Tracking

Abstract

A key problem in visual tracking is to represent the appearance of an object in a way that is robust to visual changes. To attain this robustness, increasingly complex models are used to capture appearance variations. However, such models can be difficult to maintain accurately and efficiently. In this paper, we propose a visual tracker in which objects are represented by compact and discriminative binary codes. This representation can be processed very efficiently, and is capable of effectively fusing information from multiple cues. An incremental discriminative learner is then used to construct an appearance model that optimally separates the object from its surrounds. Furthermore, we design a hypergraph propagation method to capture the contextual information on samples, which further improves the tracking accuracy. Experimental results on challenging videos demonstrate the effectiveness and robustness of the proposed tracker.

Transfer Learning Based Visual Tracking with Gaussian Processes Regression

“…Single target visual tracking has long been attracting large amounts of research efforts [39]. It is impractical to enumerate all previous work, instead we sample some recent interests related to our work: i) linear representation with a dictionary, e.g., a set of basis vectors based on subspace learning [29,12] or least softthreshold squares linear regression [32], a series of raw pixel templates based on sparse coding [25,24,44,43,36] or non-sparse linear representation [22]; ii) collaboration of multiple tracking models, e.g., Interacting Markov Chain Monto Carlo (MCMC) based [17,18,19], local/global combination based [45]; iii) part-based models, e.g., fragments voting based [1,9,5], incorporating spatial constraints between the parts [42,37], alignment-pooling across the local patches [14]; iv) and the widely followed trackingby-detection (or discriminative) methods [6,7,20,2,8,21,31,45], which treat the tracking problem as a classification task. All these trackers adaptively update tracking models to accommodate the appearance changes and new information during tracking.…”

Evsbe: Extended Visual State Binary Embedding Model for Efficient, Scalable and Fast Video Event Retrieval

“…Caffe: An open source convolutional architecture for fast feature embedding [5] Authors: Y. Jia Description: Caffe provides transmission scientists and practitioners with a clean and modifiable framework for progressive deep learning algorithms and a set of reference models. The framework may be a BSD-licensed C++ library with Python and MATLAB bindings for coaching and deploying general purpose convolutional neural networks and alternative deep models expeditiously on artifact architectures.…”

Deep Discrete Hashing with Self-supervised Pairwise Labels

“…In [40], they develop a so-called correlation component manifold space learning (CCMSL) to learn a common feature space by capturing the correlations between the heterogeneous databases. Many attempts [25,21] were focusing on compacting such high quality floating-point descriptors for reducing computation time and memory usage as well as improving the matching efficiency. In those methods, the floating-point descriptor construction procedure is independent of the hash codes learning and still costs a massive amounts of time-consuming computation.…”