Has anybody heard about openCV being used to recognize ships and other vessels in open water. Doesn't need to actually recognize the type of ship, just recognize that a shape probably IS a ship. Ideally via a fairly wide-angle field of view, with ships potentially out to the horizon.
Am just getting started on this, so any thoughts/ramblings/advice would be greatly appreciated!
In a quick analysis, I think that detecting boats and vessels in an open water scenery would be easy using a learn-based detecting algorithm. Since the background is always very similar and with almost no noise, an object like a boat will contrast a lot with the surrounding background, which is the ideal situation for these kind of algorithms.
OpenCV offers implementation of some of these algorithms, such as Viola & Jones work, widely known for being successfully applied to face detection. There are other implementations, such as HOG classification with a SVM (Support Vector Machine) machine-learning algorithm, AdaBoost feature evaluator and others.
Which solution to choose depends on the demands of your problem. Is performance as important as detection accuracy? Is it ok to miss a small percentage of objects? How much false-positives are admissible? etc.
The downside of this approach is that you need to possess a large image database of the objects you want to detect on the background the application is going to be performing on.
Look at the examples for pedestrian detection and face detection/recognition. From there you can retrain some of the learning algorithms with ships DB ;)
Asked: 2013-11-17 15:05:18 -0600
Seen: 1,308 times
Last updated: Nov 18 '13
Hello randron!
Did you solve your problem?
I am very interested in the same question.
Could I contact you privately?