Visual detection in omnidirectional view sensors

In recent years, the use of omnidirectional view (OV) sensors has gained popularity in robotics. The main reason behind this growth is due to the large field of view (FOV) that spans offered by these sensors under a catadioptric configuration. The large FOV addresses several shortcomings of a conventional perspective imaging sensor by allowing simultaneous monitoring of surrounding environment under a single image compilation. Feature detection is one of the fundamental components in visual robotics applications that enable intelligent vision system with advanced features such as object, scene, and human detection, localisation, simultaneous localisation and mapping, and odometry. In this paper, the adaptation of visual detection algorithm in omnidirectional vision is reviewed by investigating the recent works and the underlying supporting mechanism. Furthermore, state-of-the-art vision detection algorithms and important factors of OV sensors, such as hardware requirements, fundamental theories, cost, and usability, are also investigated in order to explain the adaptation involved. To conclude this work, a case study related to OV mapping transform is presented, and insights on possible future research direction are provided.