Drone Based Crowd Density Estimation and Localization Using Temporal and Location Sensitive Fused Attention Model on Pyramid Features

Abstract

Crowd monitoring is essential for security and effective management in public space, and drone imagery offers a powerful tool for this purpose. Though traditional methods often fall short in accuracy and efficiency techniques like manual counting, detection based approaches struggle with challenges like occlusion, low resolution, and high crowd density, leading to unreliable estimates. To address data privacy concerns related to capturing images of individuals without consent, regulatory barriers that restrict flight zones and operational guidelines, and technical limitations such as limited battery life and communication range like limitations, this study introduce a novel approach called the Temporal and Location Sensitive Fused Attention Model on Pyramid Features (TLFA_PF) for crowd density estimation and localization. The method employs scales while minimizing computational complexity. By integrating spatial and temporal attention schemes, the model effectively captures significant information from drone capturing images. A key innovation of this work is the introduction of a Bi Pooling Squeeze and Excitation Block, which enhances the conventional neural network by incorporating two pooling networks. This block selectively emphasizing important features improving the models ability to discern crowd density variation. The TLFA_PF model demonstrates superior performance in estimating crowd density and localizing individual compared to existing methods experimental results highlights the effectiveness of TLFA_PF across various scenario, showcasing its robustness in handling different crowd densities within the fused attention framework allows for more accurate predictions, making it’s a significant advancement in drone based crowd analysis. Overall, this research contributes to the field of computer vision by providing an efficient and effective solution for real time crowd monitoring using aerial imagery.