This article consists of 10 pages and 3145 words. In order to have full access to this article, email us at thedocumentco@hotmail.co.uk.
Ref No: 31946 (PID)

Introduction and Background

Object detection has emerged as an application of machine learning in the past few years owing to its diverse range of applications and certain breakthroughs. In the academic world, many researchers are in constant pursuit of developing algorithms both in the deep learning space and the traditional algorithms space to develop applications related to self-driving, computer vision, and transportation (Zou et al., 2019). It is important to note some of the important reasons that are often associated with the rise of object detection algorithms and some of them include the rise of cheaper computation devices and the constant development of neural networks. Neural networks are often defined as algorithms that require larger amounts of data and higher processing power. Owing to Moore’s law, the underlying hardware is doubling every year while requiring the same space which means we have the same sized microchips that have double the number of transistors (Dean et al., 2018). Currently, deep learning models have been adopted by the researchers of computer vision to develop certain algorithms that can make domain-specific object detections. This chapter is concerned with the exploration of certain object detection algorithms that have been developed recently and how to compare them for choosing a single object detection method for detecting numberplates on a car automatically.

Object detection and detectors

Object detection is a relatively development that lies at the intersection of computer vision, machine learning, and image processing. The purpose of object detection is to detect instances of semantic objects belonging to a certain class These objects can exist either in a video or a picture. Some of the important categories of object detection include edge detection, text detection, and face detection. TO evaluate the performance of any object detector, the researchers have developed certain benchmarks that can be used to evaluate the accuracy as well as the performance of an object detection algorithm. Some of the examples of such evaluation methods include MS COCO, ECCV, and ImageNet. ImageNet represents an online database of labeled and organized instances of data belonging to many classes. This database has been a major influence in the development of many object detection algorithms. Based on the statistics online, this dataset contains almost 14 million images belonging to certain classes and researchers can use this database to develop their neural network architectures (Zhou et al., 2017).

Object detection is generally categorized into two types, the first one represents the single-stage detector and the second one is the two-stage detector. Single-stage detectors are generally faster with a small tradeoff of accuracy. Two-stage detectors have higher accuracy in terms of object detection, but they require higher computation powers making it almost impossible to train the algorithm on the CPU of the machine. Two-stage detectors also require many days of training. One stage detector on the other hand can be used for real-time devices owing to their faster predictions…