HRD-GKV-CCNet: A Deep Learning-based Multitask Method for Human Crowd Management

In a unified framework, we offer a multitasking method for crowd counting, classification, and localization. As the detection, classification, counting, and localization tasks are highly associated and may be performed concurrently, our model benefits from a multitask approach of encoded crowd images and then fusing them. In contrast to the somewhat more prevalent density-based techniques, our model uses point and bounding box-based annotations to precisely identify human crowd situations. We evaluate our model on recorded videos during Kumbh Mela 2021 and demonstrate that our method achieves strong results on counting, classification, and localization tasks, with MAE measures of 105.79% for crowd counting, a precision measure of 96% for crowd classification. Our exhaustive ablation experiments demonstrate the efficacy of our multitask method.