Uniform Appearance People and Crowd Monitoring Using Drones

With the rapid increase in crowd gatherings, public safety has become a crucial concern for event organizers. Consequently, the usage of modern technology to monitor crowded locations is necessary. In this work, a challenging dataset with different crowd density scenarios was collected using a camera-equipped drone. The individuals within the crowd were assigned to be dressed uniformly to emulate real challenging scenarios. Unlike the robot, the drone possesses the ability to observe the crowd while having no impact on their mobility, rendering it a more feasible solution in practical applications. The dataset emulates crowded places at different timings and different intruder behaviors. The collected data were annotated manually utilizing Computer Vision Annotation Tool (CVAT). The results show a considerable deterioration in the performance of the State-Of-The-Art (SOTA) visual object trackers which indicates a challenging dataset to track. To enhance the performance of the tracker, STARK-ST50 was fine-tuned using 30 videos for both training and testing, and the tracker has experienced a slight improvement in its performance due to the challenging nature of the dataset.