What is the probability of the event between the i-th object and the k-?

R∗t and the authors divide the event probability by |Ri∗t | (the cardinality of a set of the relative motion models used for the i-th object) to make the total sum of event probabilities along the relative motion models 〈i, j〉 ∈ Ri∗t always be 1.

What is the assignment probability of the i-th object?

the detection event is simply obtained by the event probabilities as followsoit = ∑〈i,j〉∈R∗ ∑ k Pk(E 〈i,j〉 t ), (14)where the detection event of the i-th object is consequently defined as oit ∈ {0, 1}.

Why do the authors not include the i, j relative motion model in Rt?

the authors do not include the 〈i, j〉 relative motion model in Rt because detection failures are caused by various reasons such as disappearance and occlusion.

What is the way to evaluate the accuracy of the proposed RMN algorithm?

The authors evaluate the accuracy of the proposed RMN algorithm against a conventional self motion model (SMM) in terms of distance error on the ETH dataset as shown in Fig.

What are the main features of the proposed MOT algorithm?

The authors introduce representative MOT algorithms that focus on motion models, which can be categorized based on static or dynamic camera assumptions as considered in this work.

What is the description of the proposed algorithm?

Note that the proposed algorithm does not consider very abrupt camera motions and fluctuations, but consider realistic scenarios where the camera moves at a moderate speed with some fluctuations (as in the supplementary video) where at least a few objects can be tracked with continuous detections and predicted well, e.g., the objects in the red boxes of Fig.

Why is the posterior probability decomposed with the relative motion models?

This is because that the contributions from the other objects are not equal, and the predicted states from less contributive relative motion models are less reliable and less related to the k-th observation.

Open AccessProceedings Article10.1109/WACV.2015.12

Bayesian Multi-object Tracking Using Motion Context from Multiple Objects

Ju Hong Yoon, +3 more

- 05 Jan 2015

- pp 33-40

258

TL;DR: A Relative Motion Network (RMN) is constructed which consists of multiple relative motion models that describe spatial relations between objects, thereby facilitating robust prediction and data association for accurate tracking under arbitrary camera movements.

Abstract: Online multi-object tracking with a single moving camera is a challenging problem as the assumptions of 2D conventional motion models (e.g., first or second order models) in the image coordinate no longer hold because of global camera motion. In this paper, we consider motion context from multiple objects which describes the relative movement between objects and construct a Relative Motion Network (RMN) to factor out the effects of unexpected camera motion for robust tracking. The RMN consists of multiple relative motion models that describe spatial relations between objects, thereby facilitating robust prediction and data association for accurate tracking under arbitrary camera movements. The RMN can be incorporated into various multi-object tracking frameworks and we demonstrate its effectiveness with one tracking framework based on a Bayesian filter. Experiments on benchmark datasets show that online multi-object tracking performance can be better achieved by the proposed method.

AI Agents for this Paper

Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps

Most frequently asked questions

1. What contributions have the authors mentioned in the paper "Bayesian multi-object tracking using motion context from multiple objects" ?

In this paper, the authors consider motion context from multiple objects which describes the relative movement between objects and construct a Relative Motion Network ( RMN ) to factor out the effects of unexpected camera motion for robust tracking.. The RMN can be incorporated into various multi-object tracking frameworks and the authors demonstrate its effectiveness with one tracking framework based on a Bayesian filter.

2. What are the evaluation metrics used in the proposed algorithm?

For evaluation, the authors use well-known metrics which are widely used in MOT evaluation [12], which consists of Recall (correctly tracked objects over total ground truth), Precision (correctly tracked objects over total tracking results), and false positives per frame (FPF).

3. What is the role of data association in MOT?

In detection-based MOT, as each trajectory is constructed by matching multiple detected objects of the same class across frames, data association plays an essential role for robust tracking.

4. What is the importance of multi-object tracking?

Multi-object tracking (MOT) is of great importance for numerous computer vision tasks with applications such as surveillance, traffic safety, automotive driver assistance systems, and robotics.

Figure 1. Examples for effectiveness of a relative motion. Black box: ground truth, Red box: a well-tracked object, Magenta box: prediction based on the well-tracked object with the proposed RMN, Green box: prediction by conventional self-motion model. The distance error in each graph shows that the prediction based on well-tracked object with the RMN is much more accurate than the prediction based on the conventional self-motion model.

Table 1. Comparison with the SMOT (i.e., a baseline tracker) on datasets from a moving camera.

Figure 4. (a) Prediction accuracy of RMN and SMM after longterm mis-detections. (b) Prediction accuracy of RMN and SMM when objects are well tracked.

Figure 2. Prediction of object state transition based on RMN and (4): Rt = ∪3i=1Rit where R1t = {〈1, 1〉, 〈1, 2〉, 〈1, 3〉}, R2t = {〈2, 2〉, 〈2, 3〉}, R3t = {〈2, 3〉, 〈3, 3〉}. A detection event of the i-th object is defined as oit ∈ {0, 1} in (14). When the object is detected oit = 1; otherwise, oit = 0. In this example, the 1-st object is not detected ;hence, the relative motion models from the 1-st object are not included inR.

Table 2. Comparison with the state-of-the-art trackers. The trackers in gray color are an offline method. The evaluation results of the RMOT is obtained by using the detection results and ground truth which are the same as used in the state-of-the-art trackers.

Figure 5. Example of our results. As shown in (a), (b), (c), and (e), the objects pointed by the yellow arrow are occluded by the objects with the blue arrow. Due to occlusions, these objects are missing for a while, but after occlusion, the objects are correctly re-tracked using the RMN. As shown in (d), The object with the yellow arrow is blurred due to motion fluctuations, but it is tracked robustly with the RMN in spite of unreliable self motion model and severe appearance changes. crossing.

Citations

•Proceedings Article•10.1109/ICIP.2016.7533003

Simple online and realtime tracking

Alex Bewley, +4 more

- 02 Feb 2016

TL;DR: In this article, a pragmatic approach to multiple object tracking where the main focus is to associate objects efficiently for online and real-time applications is explored, where changing the detector can improve tracking by up to 18.9%.

...read moreread less

2.4K

•Proceedings Article•10.1109/ICIP.2016.7533003

Simple Online and Realtime Tracking

Alex Bewley, +4 more

- 02 Feb 2016

- arXiv: Computer Vision and Pattern Recog...

TL;DR: Despite only using a rudimentary combination of familiar techniques such as the Kalman Filter and Hungarian algorithm for the tracking components, this approach achieves an accuracy comparable to state-of-the-art online trackers.

...read moreread less

1.8K

•Proceedings Article•10.1109/ICCV.2015.534

Learning to Track: Online Multi-object Tracking by Decision Making

Yu Xiang, +2 more

- 07 Dec 2015

TL;DR: This work forms the online MOT problem as decision making in Markov Decision Processes (MDPs), where the lifetime of an object is modeled with a MDP, and a similarity function for data association is equivalent to learning a policy for the MDP.

...read moreread less

861

Proceedings Article•10.1109/ICCV.2015.533

Multiple Hypothesis Tracking Revisited

Chanho Kim, +3 more

- 07 Dec 2015

TL;DR: It is demonstrated that a classical MHT implementation from the 90's can come surprisingly close to the performance of state-of-the-art methods on standard benchmark datasets, and it is shown that appearance models can be learned efficiently via a regularized least squares framework.

...read moreread less

842

•Book

Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art

Joel Janai, +3 more

- 03 Jul 2020

TL;DR: This survey includes both the historically most relevant literature as well as the current state of the art on several specific topics, including recognition, reconstruction, motion estimation, tracking, scene understanding, and end-to-end learning for autonomous driving.

...read moreread less

833

...

Expand

References

•Proceedings Article•10.1109/CVPR.2005.177

Histograms of oriented gradients for human detection

Navneet Dalal, +1 more

- 20 Jun 2005

TL;DR: It is shown experimentally that grids of histograms of oriented gradient (HOG) descriptors significantly outperform existing feature sets for human detection, and the influence of each stage of the computation on performance is studied.

...read moreread less

36.7K

Journal Article•10.1023/B:VISI.0000013087.49260.FB

Robust Real-Time Face Detection

Paul A. Viola, +1 more

- 01 May 2004

- International Journal of Computer Vision

TL;DR: In this paper, a face detection framework that is capable of processing images extremely rapidly while achieving high detection rates is described. But the detection performance is limited to 15 frames per second.

...read moreread less

14.6K

Proceedings Article•10.1109/ICCV.2001.937709

Robust real-time face detection

Paul A. Viola, +1 more

- 07 Jul 2001

TL;DR: A new image representation called the “Integral Image” is introduced which allows the features used by the detector to be computed very quickly and a method for combining classifiers in a “cascade” which allows background regions of the image to be quickly discarded while spending more computation on promising face-like regions.

...read moreread less

13.8K

•Book•10.1002/0470045345

An Introduction to the Kalman Filter

Greg Welch, +1 more

- 29 Nov 1995

TL;DR: The discrete Kalman filter as mentioned in this paper is a set of mathematical equations that provides an efficient computational (recursive) means to estimate the state of a process, in a way that minimizes the mean of the squared error.

...read moreread less

4.6K

Book Chapter•10.1007/978-3-540-24670-1_3

A Boosted Particle Filter: Multitarget Detection and Tracking

Kenji Okuma, +4 more

- 11 May 2004

TL;DR: This work introduces a vision system that is capable of learning, detecting and tracking the objects of interest, and interleaving Adaboost with mixture particle filters, a simple, yet powerful and fully automatic multiple object tracking system.

...read moreread less

1.2K

...

Expand

Bayesian Multi-object Tracking Using Motion Context from Multiple Objects

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What contributions have the authors mentioned in the paper "Bayesian multi-object tracking using motion context from multiple objects" ?

2. What are the evaluation metrics used in the proposed algorithm?

3. What is the role of data association in MOT?

4. What is the importance of multi-object tracking?

5. What is the probability of the event between the i-th object and the k-?

6. What is the assignment probability of the i-th object?

7. Why do the authors not include the i, j relative motion model in Rt?

8. What is the way to evaluate the accuracy of the proposed RMN algorithm?

9. What are the main features of the proposed MOT algorithm?

10. What is the description of the proposed algorithm?

11. Why is the posterior probability decomposed with the relative motion models?

Figures

Citations

Simple online and realtime tracking

Simple Online and Realtime Tracking

Learning to Track: Online Multi-object Tracking by Decision Making

Multiple Hypothesis Tracking Revisited

Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art

References

Histograms of oriented gradients for human detection

Robust Real-Time Face Detection

Robust real-time face detection

An Introduction to the Kalman Filter

A Boosted Particle Filter: Multitarget Detection and Tracking

Related Papers (5)

Evaluating multiple object tracking performance: the CLEAR MOT metrics

Learning to Track: Online Multi-object Tracking by Decision Making

Globally-optimal greedy algorithms for tracking a variable number of objects

MOTChallenge 2015: Towards a Benchmark for Multi-Target Tracking

Learning to associate: HybridBoosted multi-target tracker for crowded scene