Shark detection in uncontrolled environment is a challenging problem that has not been paid much attention deeply. How to accomplish fast and effective coast surveillance impacts safety concerns in beach activities. This paper proposes a submerged shark detection, such as the white shark, using image identification from low altitude drones. The image identification is set on a drone to train real datasets of a 2.5 m long and 0.945 m2 shark model. The Haar feature-based cascade classifier is used to detect regions of interest (ROI) to extract some features to classify water area with a shark. The proposed system is tested in two different uncontrolled areas in Kenting coast, Taiwan by presenting contrasted conditions. The adopt technique reaches an average of 19 frames/second based on different altitudes of drone experiments from 8 to 22 m above sea level in static and dynamic detections for 30 minutes endurance. The system achieves a true detection's average of 99.5% by correct classification and the mean score on total false positive detection is 3.85%. The detection rate varies with the altitude and the weather conditions which is sensitive in building an altitude-based image detection system. The experiments show very effective results to detect sharks on sea coast to reach a lower false positive detection rate.
All Science Journal Classification (ASJC) codes
- Aerospace Engineering
- Space and Planetary Science