Robust Object Tracking with Online Multiple Instance Learning

By Boris Babenko, Ming-Hsuan Yang, Serge J. Belongie. 2011

2,097 citations recorded by the source corpus.

In this paper, we address the problem of tracking an object in a video given its location in the first frame and no other information. Recently, a class of tracking techniques called “tracking by detection” has been shown to give promising results at real-time speeds. These methods train a discriminative classifier in an online manner to separate the object from the background. This classifier bootstraps itself by using the current tracker state to extract positive and negative examples from the current frame. Slight inaccuracies in the tracker can therefore lead to incorrectly labeled training examples, which degrade the classifier and can cause drift. In this paper, we show that using Multiple Instance Learning (MIL) instead of traditional supervised learning avoids these problems and can therefore lead to a more robust tracker with fewer parameter tweaks. We propose a novel online MIL algorithm for object tracking that achieves superior results with real-time performance. We present thorough experimental results (both qualitative and quantitative) on a number of challenging video clips.

Subfields: Computer Vision and Pattern Recognition, Electrical and Electronic Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Electrical and Electronic Engineering

Citations in the corpus

Deep Perm-Set Net: Learn to predict sets with unknown permutation and cardinality using deep neural networks (2018, background): “…object detection but also all problems where a set of instances (as input or/and output) is involved, e.g. a set of topics or concepts in documents Blei et al. (2003), segmentation of object instances Lin et al. (2014) and a set of trajectories in multiple object tracking Babenko et al. (2011).”
Few-Shot Class-Incremental Learning by Sampling Multi-Phase Tasks (2022, background): “W ITH the rapid development of supervised learning, current learning systems have achieved or even sur-passed human-level performances in many tasks [1], [2], [3], [4], [5], [6].”
Learning Multi-domain Convolutional Neural Networks for Visual Tracking (2015, background): “These tracking algorithms typically learn classiﬁers based on multiple instance learning [2], P-N learning [24], online boosting [14, 15], structured output SVMs [17], etc.”
Transferring Rich Feature Hierarchies for Robust Visual Tracking (2015, methodology): “Many advanced machine learning algorithms have been used, including boosting variants [10, 11], multiple-instance learning [2], structured output SVM [12], and Gaussian process regression [7].”
In defense of color-based model-free tracking (2015, methodology): “Due to the success of discriminative models, a large variety of suitable learning methods has been explored for visual tracking, such as structured output SVMs [21], ranking SVMs [5], boosting [3, 18], kernel ridge regression [22, 23], and multipleinstance learning [4].”
PCL: Proposal Cluster Learning for Weakly Supervised Object Detection (2018, methodology): “image classification [26], [27], weakly supervised semantic segmentation [28], [29], object detection [30], object tracking [31], etc.”
Kernel Cross-Correlator (2017, methodology): “Along with the established discriminative trackers which use advanced off-the-shelf machine learning algorithms such as multiple-instance boosting (Babenko, Yang, and Belongie 2011), kernelized structured SVM (Hare, Saf-fari, and Torr 2011), graph learning (Li et al. 2017), self-supervised learning…”
Online Multi-object Tracking Using CNN-Based Single Object Tracker with Spatial-Temporal Attention Mechanism (2017, background): “Most state-ofthe-art single object tracking methods aim to online learn a strong discriminative appearance model and use it to find the location of the target within a search area in next frame [1, 17, 18, 9].”
Collective Entity Disambiguation with Structured Gradient Tree Boosting (2018, methodology): “Structured gradient tree boosting Gradient tree boosting has been used in some of the most accurate systems for a variety of classification and regression problems (Babenko et al., 2011; Wu et al., 2010; Yamada et al., 2016). However, gradient tree boosting is seldom studied in the context of struct”
Learning to Track: Online Multi-object Tracking by Decision Making (2015, methodology): “This decision making is related to the goal of single object tracking in the literature [3, 15, 16, 5].”
Video Processing From Electro-Optical Sensors for Object Detection and Tracking in a Maritime Environment: A Survey (2016, methodology): “with support for multiple instance [197]), TLD (trackinglearning-detection, a semi-supervised machine learning approach [213]), MedFlow (an optical flow technique, [214]), and KCF (a color based feature tracking approach, [215]).”
The Seventh Visual Object Tracking VOT2019 Challenge Results (2019, methodology): “MIL tracker [1] uses a tracking-by-detection approach, more specifically Multiple Instance Learning instead of traditional supervised learning methods and shows improved robustness to inaccuracies of the tracker and to incorrectly labelled training samples.”
Struck: Structured Output Tracking with Kernels (2016, positive): “…in §4 that this approach can be straightforwardly extended to deal with other types of motion, e.g. scale can be incorporated by using parameter vectors of the form (x, y, s), where s ∈ R+ denotes a factor that can be used to scale the bounding box around the object, and choosing Y appropriately.”
The Visual Object Tracking VOT2017 Challenge Results (2017, methodology): “Submitted by VOT Committee MIL tracker [1] uses a tracking-by-detection approach, more specifically Multiple Instance Learning instead of traditional supervised learning methods and shows improved robustness to inaccuracies of the tracker and to incorrectly labelled training samples.”
Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers (2018, background): “Pioneered by [26], many learning techniques have been suggested, e.g. multiple instance learning [3], structured output SVMs [19], online boosting [18], and model ensembles [4].”
Learning a Deep Compact Image Representation for Visual Tracking (2013, methodology): “Only MIL and TLD fail in some frames.”
Understanding and Diagnosing Visual Tracking Systems (2015, methodology): “Thanks to advances made by machine learning researchers, many sophisticated techniques have been applied to visual tracking, including boosting [12, 13], multiple-instance learning [3], structured output SVM [14], Gaussian process regression [11], and deep learning [36, 34, 16].”