Natural disasters are uncalled for and timely response is necessary to alleviate large damages. There are now exponentially large multimodal datasets (such as images, videos, blogs, and tweets) available in the public domain on social networking sites and online content-sharing platforms for data usability. By including local groups in data gathering and information exchange during natural disasters, it may be possible to establish data-driven approaches to damage assessment and recovery planning as well as a better understanding of flood vulnerabilities, capacities, and risks. We used state-of-the-art real-time YOLOv7 object detection for instantaneous human (and vehicles) recognition and tracking based on video content analysis of feeds. The YOLOv7 object detection pretrained model, efficient for localization and recognition in the COCO Human dataset (for training) was tested. We also custom-trained the YOLOv7 instance segmentation model in segmenting flooded areas for testing damage assessment in real time. It was trained on the River Flooding Detection System dataset, part of a flooding alert system research published in 2022, supported by the São Paulo Research Foundation (FAPESP). With the dataset, the YOLOv7 algorithm was trained and tested with a transfer learning approach on Google Colab’s virtual machine. Experimental results revealed that the YOLOv7 model achieved the Precision, Recall, and mAP@0.50 values for the bounding boxes and masks of vehicles as 0.9985, 1, and 0.995, respectively.