CNN Accelerator with FPGA: A High Performance and Energy Efficient Approach for Real Time Object Detection

Object recognition and classification problems in a variety of domains have been transformed by Convolutional Neural Networks (CNNs), particularly in autonomous vehicles and other real-time systems. CNNs are less appropriate for low-power, high-performance applications because to their high-power consumption and latency when implemented on conventional hardware architectures like CPUs and GPUs. The CNN accelerator solution we show in this study is based on an FPGA (Field Programmable Gate Array), which provides high computational throughput, flexibility, and low power consumption. At the Register-Transfer Level (RTL), low-power approaches were implemented and the design was tuned for effective data flow. When compared to traditional GPU implementations, experimental results showed notable power consumption reductions and performance gains.