Tracking Pedestrian Heads in Dense Crowd (2103.13516v1)

Abstract: Tracking humans in crowded video sequences is an important constituent of visual scene understanding. Increasing crowd density challenges visibility of humans, limiting the scalability of existing pedestrian trackers to higher crowd densities. For that reason, we propose to revitalize head tracking with Crowd of Heads Dataset (CroHD), consisting of 9 sequences of 11,463 frames with over 2,276,838 heads and 5,230 tracks annotated in diverse scenes. For evaluation, we proposed a new metric, IDEucl, to measure an algorithm's efficacy in preserving a unique identity for the longest stretch in image coordinate space, thus building a correspondence between pedestrian crowd motion and the performance of a tracking algorithm. Moreover, we also propose a new head detector, HeadHunter, which is designed for small head detection in crowded scenes. We extend HeadHunter with a Particle Filter and a color histogram based re-identification module for head tracking. To establish this as a strong baseline, we compare our tracker with existing state-of-the-art pedestrian trackers on CroHD and demonstrate superiority, especially in identity preserving tracking metrics. With a light-weight head detector and a tracker which is efficient at identity preservation, we believe our contributions will serve useful in advancement of pedestrian tracking in dense crowds.

• The paper introduces CroHD, a benchmark dataset with over 2 million head annotations to advance tracking in dense crowds.
• The paper presents HeadHunter-T, a novel tracker employing a Particle Filter and color histogram re-identification to improve tracking performance.
• The paper proposes the IDEucl metric to better evaluate tracking consistency over pedestrian trajectories compared to traditional methods.

Tracking Pedestrian Heads in Dense Crowds: A Detailed Exploration

The paper "Tracking Pedestrian Heads in Dense Crowd" by Ramana Sundararaman et al. addresses a critical challenge in computer vision: tracking individuals within dense crowds. As crowd density increases, the visibility of pedestrians diminishes due to occlusion, hampering traditional pedestrian trackers' effectiveness. This paper introduces innovative solutions, including the creation of a new dataset, CroHD, and a specialized head tracking method, which are pivotal in enhancing performance in these demanding scenarios.

Introduction and Motivation

Tracking multiple objects, particularly humans, is a fundamental problem in visual scene understanding. The complexity of this task amplifies with increasing crowd density, primarily due to frequent occlusions. Traditional pedestrian detection methods that focus on the entire body are inadequate in such settings. This paper proposes to shift the focus from full-body detection to head detection, which is less prone to occlusion and more visible in crowded scenes.

Contributions

1. CroHD Dataset: The authors introduce the Crowd of Heads Dataset (CroHD), comprising 11,463 frames with over 2 million annotated head instances across diverse scenes. CroHD targets the head detection niche by offering a benchmark dataset focused specifically on tracking human heads rather than entire bodies. This dataset is expected to stimulate further research in dense crowd tracking.
2. Head Detection and Tracking Methods: The paper introduces two primary components: HeadHunter, a novel head detector optimized for the challenges of detecting small-scale heads in crowded environments, and HeadHunter-T, a head tracker that extends HeadHunter with a Particle Filter and a color histogram-based re-identification module for enhanced identity maintenance.
3. IDEucl Metric: A new evaluation metric, IDEucl, is proposed to assess the efficacy of tracking algorithms in maintaining consistent identities over pedestrian trajectories in the image coordinate space. This metric emphasizes the distance over which consistent tracking is maintained, offering a more nuanced evaluation than existing metrics.

Results and Comparisons

The proposed HeadHunter-T tracker outperforms state-of-the-art pedestrian tracking methods on the CroHD dataset by effectively maintaining pedestrian identities over time, as indicated by superior IDEucl scores. The inclusion of a Particle Filter allows HeadHunter-T to handle non-linear motion more effectively than methods relying solely on Kalman Filters.

Implications and Future Directions

The introduction of CroHD and the accompanying methods has several implications for both theoretical and practical advancements in the field of computational vision:

• Real-time Tracking: HeadHunter-T's design, with its emphasis on maintaining identity over frames, makes it particularly suitable for real-time applications such as surveillance and crowd control in environments with high pedestrian density.
• Research Catalyst: The publicly available CroHD dataset provides a new standard for evaluating head detection and tracking algorithms, potentially leading to the development of more sophisticated techniques.
• Extended Applications: Beyond pedestrian tracking, these methods could inform advances in automated crowd analysis, anomaly detection, and even assist in developing AI systems capable of navigating through crowded environments.

Conclusion

By focusing on the visible and distinct part of the pedestrian, the head, the paper strategically addresses the key limitations faced by traditional pedestrian detectors in crowded settings. The creation of CroHD, coupled with the innovative HeadHunter and HeadHunter-T methods, marks a significant step forward in crowd tracking research. The introduction of the IDEucl metric further enriches the evaluation toolkit available to researchers, fostering more robust and reliable tracking systems. As such, this work sets a foundation for future advancements in dense crowd scene understanding.