Accuracy and efficiency trade-offs in deep learning approaches for object recognition: A comparative study

Object recognition is a research area in Computer Vision and Image Processing that covers several topics, including face recognition, gesture recognition, human gait recognition, and traffic road signs recognition, among others. It plays a vital role in several real-time applications such as video surveillance, traffic analysis, security systems, and content-based image retrieval. It is the task in which an object can be recognized and labelled within an image. It aims to bring the visual perception capabilities of human beings into machines and computers. Artificial Intelligence (AI) has a great interest in this field and is involved in most, if not all, of the various fields of life. Deep Learning (a new area of Machine Learning) is one of the AI techniques that is generated from Artificial Neural Network (ANN). Recently, it has gained popularity due to its competitive results in improving the efficiency of real-time decision-making. For this reason, it has been widely used in many fields, including object recognition. Therefore, this review provides insight into previous studies on how deep learning methods have been applied to object recognition, among various datasets. It classifies the deep learning methods along with different targeted objects, their contributions, the challenges they faced, and how the results were gained. These aspects will be discussed to introduce the researchers to more general knowledge about the recent techniques applied in this field. Practical experiments have proven the efficiency of Convolutional Neural Networks (CNNs) in object recognition tasks, achieving high-accuracy results within an acceptable timeframe.