Flag semaphore detection using tensorflow and opencv

Abstract

This paper studies the automatic recognition of Flag Semaphores. We consider both static semaphores wherein the flags are held by the signaler in fixed positions and also dynamic signaling with flags (used internationally for aircraft marshalling and also by mariners). Reading Static semaphores such as those used by mariners are our main focus. We suggest the use of image processing and machine learning techniques to recognize and detect the flags and the signaler. OpenCV technology was used to capture the images and the TensorFlow API to detect the static semaphores. We could achieve promising results in the detection of static flag semaphores - a confidence level of 99%. We conclude that for deciphering signals where the flags are in motion, more sophisticated machine learning methods would be needed.