Crowd detection and counting are essential in public safety, event management, and urban planning applications. High population density poses unique challenges due to overlapping individuals, occlusions, and varying crowd densities. This study compares four You Only Looks Once (Yolo) object detection models such as Yolov5, Yolov7, Yolov8, Yolov9, and Yolov11 to evaluate their effectiveness in detecting and counting individuals in sparse settings. The models are trained and tested on the two distinct datasets. Results show that Yolov7 and Yolov11 outperform the other Yolo models. These findings highlight the advantages of Yolov7 and Yolov11 for accurate crowd detection in real-world scenarios and underscore the importance of model selection in crowd analytics.