| Summary: | 碩士 === 國立臺北科技大學 === 電子工程系 === 107 === Typical advanced driving assistance systems (ADAS) are popular and growing. The concept of personalized adaptive cruise control (PACC) is to keep comfort and safety. Recently, machine learning is adopted to find the relationship between driver’s behavior and adjusted distance. On related research proposed a PACC system using deep neural network (DNN) training method. However, DNN is a computationally-intensive method, its high computational complexities make power consumption increase more may be not suitable for embedded systems. In this work, we proposed a hardware DNN architecture by considering the high accuracy and low power consumption. The major focus of this thesis is to implement car-spacing learning module (training) and prediction module with edge computing by hardware. The proposed method is implemented on Xilinx Zedboard. Experimental results show that the proposed hardware architecture improves computationally intensive problems, and keeps high-accuracy computation and reduces on power consumption by 27% than the other GPU architecture.
|
|---|
