An efficient real-time FPGA implementation for object detection

In this paper, we present an efficient real-time FPGA implementation for object detection. The system employs Speeded Up Robust Features (SURF) algorithm to detect keypoints on every video frame and applies Fast Retina Keypoint (FREAK) method to describe the keypoints. One-to-one feature matching is performed between the descriptors of objects in the library and the descriptors of the video frames, to ensure a high object detection accuracy. Our experiments demonstrate that our FPGA-based design is fully functional and it can process video frames with 800×600 resolution at 60 fps. The proposed FPGA design is 23 times faster than the same algorithm implemented on Intel Core i5-3210M CPU.