FPGA based Feature Extraction in Real time Computer Vision-A Comprehensive Survey

Today, Computer Vision algorithms play a vital role in almost every domain of our day-to-day life. This powerful technology has aided in the development of answers to a wide range of difficult and complex challenges encountered across different sectors like defence, security, surveillance, robotics, healthcare, transportation, manufacturing, agriculture, sports, retail etc. Feature Extraction is one of the key components in any computer vision system and it aims at extracting the distinct and relevant features from an image that could clearly characterize different objects present in it. Almost every computer vision application like image matching, object detection and recognition, human or pedestrian detection, robotic vision etc. rely on the feature extraction techniques as one of their primary steps. Recently various hardware accelerators have been developed for these feature extraction processes and among these Field Programmable Gate Array (FPGA) is considered as the most suitable platform because of its added advantages of flexibility, parallelism, low power, and re-configurability. This paper presents, a review of the literature on FPGA implementations of various efficient feature extraction methods, such as Scale Invariant Feature Transform (SIFT) and Histogram of Oriented Gradients (HoG). The review covers the different implementation techniques followed by researchers for efficient mapping on hardware and provides a comparison of their work on the basis of parameters like speed and resource utilization.