Back to top
Syna Theme

Robot-Person Tracking in Uniform Appearance Scenarios

Motivations

A Large-Scale, First of its kind Dataset, and Benchmark for Person Tracking in the presence of highly similar appearing distractors.

45 Video Sequences with several challenging attributes

Contains No distractors videos and With distractors videos

More than 85K bounding boxes

Key Features

Incorporates several real-life possible situations for which the trackers should be robustly evaluated

Uniformity

This is the unique attribute of our dataset that introduces a very specific challenge to the tracking paradigm. People, including the target and non-target persons (distractors), wear the same garment during the whole recording.

Camouflage

This feature introduces a situation in which the background is highly similar to the target appearance. The background of the environment is same as target appearance.

Static Obstacles

Static obstacles are objects placed in the scene, such as boxes and banners, which can hide the target person partially or fully when he is walking during the recording.

Pose Variation

Human body is deformable on its own, however to include different poses we have explicitly added poses like running, stretching, and squatting in the videos.

Illumination Variation

The video sequences also has the illumination intensity of the environment changed from light to dark during tracking.

Occlusion

Occlusion is caused by the integration of obstacles in the scene or by the presence of other people partially cover the appearance of the target.

Motion Blur

The blur effect has been added as a nuisance factor in the video to assess the robustness of the tracking.

Out of View

This situation corresponds to an attribute when the target person leaves the frame or is not visible in the frame. The availability of the target person cannot be known/confirmed given the information in the frame.

First frame of the collected video sequences

Performance comparison of SOTA Trackers

Two different scenarios including single-person tracking and multi-person tracking in uniform appearance scenarios are considered. And, the performance of the trackers highly degrades in the presence of uniform appearance in the scenes.

Visualization of the SOTA trackers’ performance on PTUA dataset

Recent News

  • [10 April 2023] "Robot-Person Tracking in Uniform Appearance Scenarios: A New Dataset and Challenges" published in IEEE Transactions on Human Machine Systems.
  • [09 May 2022] "Robot Person Following in Uniform Crowd Environment" accepted in ICRA 2022 Workshop.
  • [04 Feb 2022] PTUG dataset submitted to IEEE Transactions on Human Machine Systems.

Publications

  • Robot-Person Tracking in Uniform Appearance Scenarios: A New Dataset and Challenges

    Xiaoxiong Zhang*, Adarsh Ghimire*, Sajid Javed, Jorge Dias, Naoufel Werghi
    IEEE Transactions on Human Machine Systems [PDF]

  • Robot Person Following in Uniform Crowd Environment

    Adarsh Ghimire*, Xiaoxiong Zhang*, Sajid Javed, Jorge Dias, Naoufel Werghi
    ICRA Workshop 2022: ROBOTIC PERCEPTION AND MAPPING: EMERGING TECHNIQUES [PDF]

OUR TEAM

Adarsh Ghimire

Adarsh Ghimire

Master Student
Xiaoxiong Zhang

Xiaoxiong Zhang

PhD Student
Sajid Javed

Sajid Javed

Assistant Professor
Jorge Dias

Jorge Dias

Full Professor
Naoufel Werghi

Naoufel Werghi

Full Professor