The explosively increasing demands of high-speed data applications have brought massive access requirements to various mobile devices. As integrating with artificial intelligence and neural network, the mobile device industry is often more concerned with faster inference with lower power consumption, bringing deep learning inference acceleration to the spotlight. In this paper, we perform a neural network inference merging R-CNN, an object detection model, into a deep learning accelerator architecture. It is a brand new implementation of neural network on embedded system hardware IP-cores of edge computation. On one hand, based on the embedded system environment, we implement the image pre-processing with region proposal algorithm and image post-processing with NMS method. On the other hand, we perform the feature computation with the deep learning accelerator through optimized software and hardware configurations. Through this method, we solve the problem of time-consuming in the computation of neural network layers and give a precise and real-time prediction of object detection. Our R-CNN inference achieves impressive results with 1.9 to 2.6 times higher performance compared with other inference processors.