People analytics in crowded environments is important for public safety, event management, and urban planning. Anyway, traditional methods typically fail in high density crowds, due to occlusions and low-resolution people to be counted. This is much more true in case the videos are acquired by a camera mounted on board of a drone. Within this context, we propose a novel approach combining point-based formulations with ConvNeXt architecture and an enhanced Fine-Grained Feature Pyramid (eFGFP) for multi-scale management, including an additional layer at 1/4 resolution, improving the handling of small objects in complex scenarios. ConvNeXt enhances feature extraction and representation, while eFGFP preserves fine-grained information across different scales. Evaluated on 19 datasets with over 43,000 images, our architecture demonstrates impressive performance compared to state-of-the-art methods, offering a robust solution for crowd counting in highly crowded environments.