Object detection is widely used in fields such as intelligent surveillance and autonomous driving. Currently, object detection algorithms based on convolutional neural networks have achieved significant performance improvements. However, due to the complexity of the algorithms and computational constraints, it is challenging to deploy them on edge computing platforms to achieve near sensor intelligence. Therefore, we have optimized and quantized the lightweight Yolov5s model to obtain a hardware-friendly Q-Yolov5s. We propose a high-performance accelerator based on the hybrid streaming architecture. The experimental results on the AX7350 FPGA show that the throughput and energy efficiency of the accelerator are 10.80 GOPs and 78.62 Pixels/mJ, respectively. Compared with the existing work, the increases are 85.25% and 35.41%, respectively. And the energy efficiency of the accelerator is 2.0 and 2.2 times higher than that of Intel i7-12700 CPU and NVDIA RTX 3070 GPU, respectively. Therefore, it is more suitable for deploying on edge computing platforms to achieve near sensor intelligence.