Unmanned Aerial Vehicles for Crowd Surveillance

Detecting the crowd plays a pivotal role as the current population of India has crossed 140 crores and was expected to reach 166 crores by 2050. Detection and analysis of crowds help in ensuring safety, mitigating traffic congestion, preventing accidents etc. The current study explores the effectiveness of the object detection algorithms, namely SSD MobileNetv3 and state-of-the-art YOLO algorithms, in the context of crowd detection. The proposed approach is tested on MSCOCO Dataset and on the custom dataset which were annotated in YOLO format that were collected various parts of the country. Each algorithm was evaluated based on several key performance indicators, including the processing speed, precision, accuracy etc. Furthermore, we investigated various hyperparameters such as input resolution, training epochs to increase the performance of the software. The resulting software was integrated into a Drone, an Unmanned Aerial Vehicle (UAV) that helps in surveillance for detecting crowd and provide insights related to the crowd prediction in various parts of a city through a dashboard that send alerts to the concerned department to predict and forecast the crowd.