Real Time Tracking-With-Detection for Coping with View Point Change

Shaul Oron, Aharon Bar-Hillel, Shai Avidan


Abstract

We consider real-time visual tracking with targets undergoing view-point changes. The problem is evaluated on a new and extensive dataset of vehicles undergoing large view-point changes. We propose an evaluation method in which tracking accuracy is measured under real-time computational complexity constraints and find that state-of-theart agnostic trackers, as well as class detectors, are still struggling with this task. We study tracking schemes fusing realtime agnostic trackers with a non-real-time class detector used for template update, with two dominating update strategies emerging. We rigorously analyze the template update latency and demonstrate such methods significantly outperform stand-alone trackers and class detectors. Results are demonstrated using two different trackers and a state-ofthe- art classifier, and at several operating points of algorithm/ hardware computational speed.

Publications

    Real Time Tracking-With-Detection for Coping with View Point Change PDF BIB

    Shaul Oron, Aharon Bar-Hillel, Shai Avidan
    Machine Vision and Application (MVAP) 2015

Code

Comming up soon...

Tracking Results

Tracking-with-detection methods in orange.

2044

3054

3064

3069


Dataset and Results

Results for all methods as reported in the paper

Option 1 (signle file 10GB)
Vehicle video dataset (zip file 10GB)

Option 2 (3 files 4.6GB, 4.6GB, 2GB):
Vehicle video dataset (tar file 4.6GB)
Vehicle video dataset (tar file 4.6GB)
Vehicle video dataset (tar file 2GB)
Must download all 3 files before extracting archive

If using this dataset please cite these papers:
[1] Real Time Tracking-With-Detection for Coping with View Point Change BIB
Shaul Oron, Aharon Bar-Hillel, Shai Avidan
Machine Vision and Application (MVAP) 2015

[2] Extended Lucas-Kanade Tracking BIB Website PDF
Shaul Oron, Aharon Bar-Hillel, Shai Avidan.
European Conference on Computer Vision (ECCV) 2014.

The dataset contains 166 sequences recorded using several cameras installed on a maneuvering vehicle facing back.
The sequences focus on vehicles undergoing view point changes in scenarios such as turns, traffic circles and overtakes.
The data is fully annotated including a bounding box and rough viewpoint estimation per frame.
For more information see the attached README file.
Website