A Systematic Review on Object Localisation Methods in Images

Abstract

Currently, many applications require a precise localization of the objects that appear in an image, to later process them. This is the case of visual inspection in the industry, computer-aided clinical diagnostic systems, the obstacle detection in vehicles or in robots, among others. However, several factors such as the quality of the image and the appearance of the objects to be detected make this automatic location difficult. In this article, we carry out a systematic revision of the main methods used to locate objects by considering since the methods based on sliding windows, as the detector proposed by Viola and Jones, until the current methods that use deep learning networks, such as Faster-RCNN or Mask-RCNN. For each proposal, we describe the relevant details, considering their advantages and disadvantages, as well as the main applications of these methods in various areas. This paper aims to provide a clean and condensed review of the state of the art of these techniques, their usefulness and their implementations in order to facilitate their knowledge and use by any researcher that requires locating objects in digital images. We conclude this work by summarizing the main ideas presented and discussing the future trends of these methods.

Keywords

• Algoritmos de detección
• Aprendizaje máquina
• Procesamiento de imágenes
• Reconocimiento de objetos
• Reconocimiento de patrones