Abstract:
Convolutional Neural Network (CNN) is an in-depth learning algorithm widely use in image processing and pattern recognition due to its robustness that includes flexibility. However, it is also the most calculated that leads to real-time malpractice. Field Programmable Gate Arrays (FPGA) works well for power consumption, flexible processing flexibility and pipeline performance. It is therefore expected to be used to accelerate an in-depth learning algorithm. In this study, a FPGA-based system was developed to detect hand-held visual perception in real time. We train the CNN model caffeinated model and get the model parameters on the PC. The Bilinear interpolation algorithm is used to adjust the image size taken by the camera. We then use FPGA to perform the Guessing Hand Gesture Recognition process and the parameters obtained by designing the accelerator using Xilinx tools.
