With the rapid development of 5G and artificial intelligence, autonomous driving and smart transportation have attracted more and more attention. Traditional target detection algorithms have deep network layers and a large number of parameters, and a lot of time is required for model detection, which seriously affects the real-time performance of the detection system. The Mobilenet_SSD network uses deep separable convolution to reduce model parameters, thereby shortening computing time. In this paper, we use the Mobilenet_SSD model and optimize the frozen network structure using the openvino tool and combine it with asynchronous operations for computation. Experimental results show that the use of Mobilenet_SSD in combination with the openvino tool and asynchronous operation techniques further improves the detection performance of the system.