In mission-critical applications such as disaster response and disaster management in response to natural calamities, every country has a rescue plan to act upon and minimize the life and property damages. However, deploying rescue services to scan large neighborhoods through ground vehicles is time-consuming and may potentially put the rescue team at risk. AI (artificial intelligence)-assisted detection and analysis with drone imagery can provide an efficient and faster solution. The primary limitation of using A.I. techniques is the unavailability of curated datasets for AI model training for AI-based interventions to accurately detect and spot objects and develop situational awareness for timely, efficient search and rescue. To address this, we present a novel dataset of images captured by a drone consisting of two target classes (humans and vehicles) collected and manually annotated from the actual drone video footage. The data is diverse across disaster types and geographic locations with generalization capabilities. We present the preliminary results of the Roboflow 3.0 object detection (fast) model trained on our dataset. The data can be further leveraged to train and deploy deep learning-based object detection models for efficient terrain scanning without costing human capital.