Workshop program

Session 1: The Visual Object Tracking VOT2013 challenge results Presentation Paper

Matej Kristan (University of Ljubljana), Roman Pflugfelder (Austrian Institute of Technology), Aleš Leonardis (University of Birmingham), Jiri Matas (Czech Technical University in Prague), Fatih Porikli (Mitsubishi Electric Research Laboratories), Luka Cehovin (University of Ljubljana), Georg Nebehay (Austrian Institute of Technology), Gustavo Fernandez (Austrian Institute of Technology), Tomas Vojir (Czech Technical University in Prague), et al.

Coffee break

Session 2: Tracker Presentations

Abstract: Tracking by detection techniques have recently been gaining popularity and showing promising results. They use samples classified in previous frames to detect an object in a new frame. However, because they rely on self updating, such techniques are prone to object drift. Multiple classifier systems can be used to improve the detection over that of a single classifier. However, such techniques can be slow as they combine information from different tracking methods. In this paper we propose a novel real-time ensemble approach to tracking by detection. We create a diverse ensemble using random projections to select strong and diverse sets of compressed features. We show that our proposed ensemble tracker significantly improves the accuracy of tracking while not using any additional information than that available to the single classifier; thus requiring little extra computational overhead. Our results also show that employing our multiple classifier system with feature subsets gives significantly better results than directly combining the features.

Abstract: Visual tracking of objects under varying lighting conditions and changes of the object appearance, such as articulation and change of aspect, is a challenging problem. Due to its robustness and speed, distribution field tracking is among the state-of-the-art approaches for tracking objects with constant size in grayscale sequences. In the present paper we use the theoretic connection between averaged histograms and channel representations to derive an enhanced computational scheme. This enhanced distribution field tracking method outperforms the state-of-the-art method in all three aspects of the VOT evaluation: accuracy, robustness, and speed.

Coffee break

Session 3: Tracker Presentations

12:00
An Adaptive Combination of Multiple Features for Robust Tracking in Real Scene
Weihua Chen, Lijun Cao, Junge Zhang, Kaiqi Huang (Chinese Academy of Sciences)

Abstract: Real scene video surveillance always involves low resolutions, lack of illumination or cluttered environments, which leads to insufficiency of discriminative details for the target. In this situation, discrimination based tracking methods could fail. To address this problem, this paper presents an adaptive multi-feature integration method in terms of feature invariance, which can evaluate the stability of features in sequential frames. The adaptive integrated feature (AIF) is consisted of several features with dynamic weights, which describe the degree of invariance of each single feature. An incremental principal component analysis (IPCA) adjusted by the accuracy of tracking results is used to update the adaptive integrated feature, and partially avoids the problem of 'updating dilemma', which is common in most of adaptive updating methods. Experiments on pedestrian tracking demonstrate the proposed approach is effective and shows improved performance compared with several state-of-the-art methods in real surveillance scenes.

Abstract: This paper addresses the problems of tracking targets which undergo rapid and significant appearance changes. Our starting point is a successful, state-of-the-art tracker based on an adaptive coupled-layer visual model. In this paper, we identify four important cases when the original tracker often fails: significant scale changes, environment clutter, and failures due to occlusion and rapid disordered movement. We suggest four new enhancements to solve these problems: we adapt the scale of the patches in addition to adapting the bounding box; marginal patch distributions are used to solve patch drifting in environment clutter; a memory is added and used to assist recovery from occlusion; situations where the tracker may lose the target are automatically detected, and a particle filter is substituted for the Kalman filter to help recover the target. We have evaluated the enhanced tracker on a publicly available dataset of 16 challenging video sequences, using a test toolkit. We demonstrate the advantages of the enhanced tracker over the original tracker, as well as several other state-of-the art trackers from the literature.

Lunch

Coffee break

Session 5: Tracker Presentations, Discussion

16:00
Graph Embedding Based Semi-Supervised Discriminative Tracker
Jin Gao, Junliang Xing, Weiming Hu, Xiaoqin Zhang (Chinese Academy of Sciences)

Abstract: Recently, constructing a good graph to represent data structures is widely used in machine learning based applications. Some existing trackers have adopted graph construction based classifiers for tracking. However, their graph structures are not effective to characterize the interclass separability and multi-model sample distribution, both of which are very important to successful tracking. In this paper, we propose to use a new graph structure to improve tracking performance without the assistance of learning object subspace generatively as previous work did. Meanwhile, considering the test samples deviate from the distribution of the training samples in tracking applications, we formulate the discriminative learning process, to avoid overfitting, in a semi-supervised fashion as 1-graph based regularizer. In addition, a non-linear variant is extended to adapt to multi-modal sample distribution. Experimental results demonstrate the superior properties of the proposed tracker.

Abstract: Long term tracking of an object, given only a single instance in an initial frame, remains an open problem. We propose a visual tracking algorithm, robust to many of the difficulties which often occur in real-world scenes. Correspondences of edge-based features are used, to overcome the reliance on the texture of the tracked object and improve invariance to lighting. Furthermore we address long-term stability, enabling the tracker to recover from drift and to provide redetection following object disappearance or occlusion. The two-module principle is similar to the successful state-of-the-art long-term TLD tracker, however our approach extends to cases of low-textured objects.
Besides reporting our results on the VOT Challenge dataset, we perform two additional experiments. Firstly, results on short-term sequences show the performance of tracking challenging objects which represent failure cases for competing state-of-the-art approaches. Secondly, long sequences are tracked, including one of almost 30 000 frames which to our knowledge is the longest tracking sequence reported to date. This tests the re-detection and drift resistance properties of the tracker. All the results are comparable to the state-of-the-art on sequences with textured objects and superior on non-textured objects. The new annotated sequences are made publicly available.

• David Liebowitz, General Dynamics Mediaware
• Mubarak Shah, University of Central Florida
• Jiri Matas, University Prague
• Federico Pernici, University Florence
• Navid Nourani-Vatani, CSIRO

• Which evaluation criterias: speed v. accuracy and reliability?
• How to design a standardised evaluation kit?
• Future challenges on particular applications?

Closing, Concluding remarks