Crown Detection in Yakshagana Images Using SSD

Abstract

A computer vision technique called object detection is used to recognise and locate items in an image or video. We make use of bounding boxes which allow us to know where the object is located in an image or a video. After locating the objects, we use different methods to group them into categories and know what the detected object is. Object detection is used in various areas like self-driving car, number plate recognition, video surveillance etc. Different methods for detecting objects have been developed in the past years. Main categories of object detection techniques are traditional computer vision and deep learning. Here in this article, we discuss about object detection using SSD and also know its advantages when compared to other object detection algorithms. We walk through the different steps involved in object detection using SSD. We make use of SSD where the output of the bounding boxes is divided into a set of boxes by default having different sizes and aspect ratios. As seen on the feature map, the network creates modifications to the default boxes to better fit the form of the objects and scores are determined by the number of times an item category occurs inside each box during prediction. We use several feature maps having different resolutions and the output obtained are combined together to help manage objects of different sizes. Further we make use of SSD for crown detection in Yakshagana images.