Affiliation:
1. Sanyo-Onoda City University, Japan
2. The American University in Cairo, Egypt
3. Okayama University, Japan
Abstract
Previous studies have successfully identified results of searching the desired object in real-time movie frames using a template image prepared in color or shape. In every sample time of 40 ms, the center of gravity (COG) can be consecutively calculated by detecting and identifying the desired object. Hence, it is possible to have visual feedback (VF) control with the quadrotor by referring to the change of the COG as the relative velocity with respect to the desired object. In this chapter, a useful function is developed to enable the detection and counting of desired objects in real-time. This allows the quadrotor to monitor the number of individually selected objects in the frames of a movie such as cars, animals, etc. This supports high-speed counting while avoiding errors due to object overlapping. Also, the proposed VF controller is successfully applied to a pick-and-place robot, in which a transfer learning-based convolutional neural network can estimate object orientation for smooth picking.