Red and green parking lamp location detection method in UAV environment

This paper suggests a method to detect locations of red and green parking lamps in UAV environment. Red and green parking lamps occupy very small space in an entire image with a very simple ellipse shape so the images go through pre-processing based on color in order to lower the frequency of false positive on parking lamps. Then, a positive sample should be created with images of parking lamps and a negative sample with a sub-image of a background without parking lamps for SVM learning. SVM will then learn the created samples and detect parking lamps through HOG detection which decides whether the sub-images of test images are parking lamps or not. The test was conducted on video clips recorded inside through a flight of an AR. Drone. Performance was analyzed with computed precision and recall values. Red lamp precision was 0.968 and recall, 0.981, thus, the performance was high whereas the values of green lamp were relatively low with precision of 0.959 and recall of 0.55.