Prevent usage of shark-nets and reduce negative shark-human interactions
1. Multiple drones deployed on flight paths to obtain video of sharks and non-sharks over several weeks in target area. Use same camera as for final detection system, and at same altitude absl. Polarising filter?
2. Label positive (sharks) and negative (e.g. dolphins, boats) objects in video (e.g. using Vatic) with bounding boxes and annotations
3. Apply feature extractor (HOG) to positive and negative samples -> giving descriptive feature vector for classifier. You can combine Colour Histogram and HOG. It would be likely to transform to HSV colour space?
4. Train classifier e.g. Support Vector Machine (SVM) with a linear function kernel using the obtained feature vector dataset. Obviously can use other classifiers.
5. Implement active learning iteration
6. Test classifier
By Air T&G over Byron Bay:
Tiger sharks at Trigg Beach, as seen from the Westpac lifesaver rescue helicopter.
So they are not always so easy to spot from the air! Although by altering contrast:
The design of the buoy-aerostat linking system. This is to supply helium to the aerostat, power the camera, and send real-time video to shore-based computers:
Aerostats could be deployed between 100-200m asl depending on compromises between field of view vs resolution. The constant helium top-up supply could allow aerostats to remain airborne for several days to weeks. Additional energy could be supplied by a wind turbine on the buoy or attached to the tether. Buoys would be anchored to the ocean floor. There may be confounds in using the camera in high winds - even with gyro stabilisation.