A Review on Traditional and Deep Learning based Object Detection Methods

Fast and accurate object detection systems are in high demand due to the advent of autonomous vehicles, smart video surveillance, facial detection, and numerous people counting applications. These systems not only detect and classify every object in an image or video, but also locate each one by creating a bounding box around it. This paper analyses the traditional and recent deep learning-based object detection methods from different perspectives, incorporating features recognition on many scales, data expansion, training approach, and perspective detection, in order to make it easier to deeply understand object detection. Some commonly used standard datasets for object detection are discussed. It also addressed the challenges and possible research scope in the future from the perspective of evolving object detection datasets and the framework for object detection tasks. From the analysis, it is observed that the performance of the methods in use for object detection is moderate and requires improvement, especially in difficult environments such as large object scale variance, obstructed object view, and horrific mild prerequisites. Therefore, the possible research scope for inventions and implementation of more novel deep learning methods to enhance object detection and classification accuracy is discussed.