Rescue Vision: A Comprehensive Video Dataset for Real-Time Flood Survivor Detection with YOLOV8

South Asian floods have emphasized the urgent need for increased survivor location and rescue efforts, since late intervention in danger zones always results into catastrophic loss of human life. To address the need, this research proposes a real-time flood survivor locator system with a large-scale custom dataset called the FloodSurvivorVision. The key characteristics of our dataset include a vast collection of diverse flood scenarios, precise annotations for survivor detection, and comprehensive coverage across varied environmental conditions. The dataset consists of around 459,660 images, and out of which around 37,230 images were utilized for training YOLOv8 model to identify real-time video streams and images. Proposed methodology also address the challenges of occlusion and varying environments through strong object detection mechanisms. The system, achieved a mean Average Precision (mAP50) of 0.926, precision of 0.877, and recall of 0.854, and demonstrated how effective it could be for real-time detection and presented a potential tool for the enhancement of flood rescue operations. This new approach, in addition to attempting to maximize survivor identification speed and accuracy, also attempts to combine new technology with practical application for disaster scenarios. Crucially, this flexible model and dataset can be utilized under various conditions of flooding as a basis for conducting further research in emergency response method.