Billboard Detection in the Wild

Advertising has a huge impact on modern life hence its analysis is very important. Billboard detection in a scene is very challenging given the outdoor position of billboards and the changing nature of a board’s size, scale, and the angle at which it is viewed by oncoming traffic or pedestrians. Hence the requirement to detect the billboard and determine the visibility to consumers is a very difficult task. In this paper, we propose a system which will not only detect a billboard but also classify the different types of billboard panels. There exists a number of different types of billboard such as Street Furniture, Roadside, in-Mall, or Spectacular to name a few, however here we focus solely on Street Furniture and Roadside. For this, the Tensorflow object detection API is used with the Single Shot Multibox Detector (SSD) architecture. SSD is chosen because of its high-speed computation and ability to eliminate false-positive cases. In this paper, we demonstrate SSD’s detection performance by fine-tuning hyperparameters and illustrate this using a dataset of billboards in the wild.