Improving trajectory continuity in drone-based crowd monitoring using a set of minimal-cost techniques and deep discriminative correlation filters

TL;DR

The paper addresses trajectory discontinuities in drone-based crowd tracking by introducing a point-oriented online tracker built on SORT, complemented with camera motion compensation, altitude-aware assignment, and a classification-based validation step. It further refines tracking with Deep Discriminative Correlation Filters that reuse spatial features from the localisation module, enabling robust, real-time performance. Evaluations on DroneCrowd and the new UP-COUNT-TRACK dataset demonstrate substantial improvements in tracking continuity and counting accuracy, with counting errors reduced to 23% and 15%, respectively, and reduced identity switches compared to baselines and competitive with offline methods. The approach offers a practical, scalable solution for drone-based crowd monitoring, with a public dataset to benchmark future developments and applications in crowd analysis and urban planning.

Abstract

Drone-based crowd monitoring is the key technology for applications in surveillance, public safety, and event management. However, maintaining tracking continuity and consistency remains a significant challenge. Traditional detection-assignment tracking methods struggle with false positives, false negatives, and frequent identity switches, leading to degraded counting accuracy and making in-depth analysis impossible. This paper introduces a point-oriented online tracking algorithm that improves trajectory continuity and counting reliability in drone-based crowd monitoring. Our method builds on the Simple Online and Real-time Tracking (SORT) framework, replacing the original bounding-box assignment with a point-distance metric. The algorithm is enhanced with three cost-effective techniques: camera motion compensation, altitude-aware assignment, and classification-based trajectory validation. Further, Deep Discriminative Correlation Filters (DDCF) that re-use spatial feature maps from localisation algorithms for increased computational efficiency through neural network resource sharing are integrated to refine object tracking by reducing noise and handling missed detections. The proposed method is evaluated on the DroneCrowd and newly shared UP-COUNT-TRACK datasets, demonstrating substantial improvements in tracking metrics, reducing counting errors to 23% and 15%, respectively. The results also indicate a significant reduction of identity switches while maintaining high tracking accuracy, outperforming baseline online trackers and even an offline greedy optimisation method.

Figures (10)

• Figure 1: An example frame demonstrates the task of localising and tracking tiny objects. Marked trajectories represent the objects' current positions and their movement history over the past two seconds.
• Figure 2: Example frames with annotated trajectories illustrate the dataset's diverse recording environments, varying flight altitudes, and dynamic drone movements.
• Figure 3: Architecture of a simple convolutional network designed to classify if a region of interest contains a person.
• Figure 4: Usage of Deep Discriminative Correlation Filters in drone-based people tracking for sample images. The heatmap indicates the locations with the highest correlation within the analysed image area.
• Figure 5: Sixteen deep features extracted in the neighbourhood of a person. Values are normalised to a zero-one range for visualisation purposes.