Overview on Interest Point Detection Over 40 Years Development: A Review

Interest point detection (IPD) is one of the most fundamental studies in computer vision and other communities. The methods for IPD have evolved from early simple mathematical model-based to current artificial intelligence (AI)-based methods, from early 2-D gray image-based to 3-D point cloud data-based methods, and from early personal computer (PC)-based to field programmable gate array (FPGA) chip-based onboard methods. Many scholars are very concerned about these methods' advantages and disadvantages and what the future development on the IPD method is; therefore, this article makes a deep and comprehensive overview for two major types of the IPD methods, the hand-crafted and machine learning, including their advantages, disadvantages, localization accuracy, noise immunity and others, for over 40 years. A comprehensive analysis using average repeatability (AR) and localization error (LE) is also evaluated. The future development for the IPD is finally given.