Correction of Omnidirectional Camera Images using Reconfigurable Hardware

The use of omnidirectional cameras in computer vision algorithms is an additional computational load if the processing algorithm needs to extract the rectangular transformed image. There exist different alternatives to deal with this task inside desktop computers but it would be desirable to design a hardware correction unit in order to operate in real-time while using minimal silicon area resources. The proposed design allows the online transformation of a circular grayscale image coming from an omnidirectional camera to obtain the corresponding rectangular image. The model parameters include the image size and radial center. The processing time is one output pixel per clock cycle (50 MHz) which allows the image to be transformed before the arrival of the next one in most applications. It is possible to insert the developed FPGA core in the digital interface of the current omnidirectional cameras