Person Monitoring by Full Body Tracking in Uniform Crowd Environment
This work addresses a specific problem for surveillance and security applications in the Middle East, but it is incremental as it focuses on dataset creation and fine-tuning rather than novel method development.
The authors tackled the problem of full body tracking in uniform crowd environments, which challenges existing trackers, by creating an annotated dataset and fine-tuning a state-of-the-art tracker, resulting in improved performance on evaluation metrics compared to the pre-trained version.
Full body trackers are utilized for surveillance and security purposes, such as person-tracking robots. In the Middle East, uniform crowd environments are the norm which challenges state-of-the-art trackers. Despite tremendous improvements in tracker technology documented in the past literature, these trackers have not been trained using a dataset that captures these environments. In this work, we develop an annotated dataset with one specific target per video in a uniform crowd environment. The dataset was generated in four different scenarios where mainly the target was moving alongside the crowd, sometimes occluding with them, and other times the camera's view of the target is blocked by the crowd for a short period. After the annotations, it was used in evaluating and fine-tuning a state-of-the-art tracker. Our results have shown that the fine-tuned tracker performed better on the evaluation dataset based on two quantitative evaluation metrics, compared to the initial pre-trained tracker.