Abstract
Rip current is a narrow but strong water current directing away from the shore. It can be nearly impossible to be spoted by the untrained eyes, accounts for more than 80 percent of the rescues performed by surf beach lifeguards and leads to many deaths every year in the United States. Oceanographers normally use a Doppler velocimeter or the millimeter-wave radar to collect wave data to detect rip currents. While this equipment has high accuracy for visible surface currents, it is too expensive to cover vast ranges of beaches and coastal areas. It requires a large amount of setup time and can be dangerous to deploy in-situ when rips are present. That is, it is difficult to have a real-time and dense estimation of the velocity of currents in the near-shore zones by utilizing the existing techniques. With the development of UAVs (Unmanned Aerial Vehicles) and camera technologies, it becomes easier to obtain high-resolution videos of near-shore areas by UAVs, which enables the near-shore area monitoring in an affordable and scalable way. However, in order to prevent the risks of rip currents, it is more desirable to develop a video analysis technique that can detect rip current automatically. In this thesis, a novel video-based rip current detection framework based on optical flow estimation and temporal data fusion techniques is proposed. Based on our empirical experiments on videos collected in Haulover and Palm Beach, our proposed framework can effectively produce the fine-grained regions of rip currents using UAV videos and the results match with the human expert's annotations. This detection system has the advantages of low cost and easy to deploy. The framework in this thesis can become a improved and extended warning system for rip currents and other hazardous beach conditions. The rip current detection system will also be potentially a useful tool for oceanographers and coastal scientists to monitor the other currents in the near-shore zones and to track the transportation of biota, pollutants, and sediments.