Vehicle detection is an important part of environmental awareness in the Advanced Driver Assistance System (ADAS). Traditional vehicle detection is usually performed on a personal computer, with the rapid improvement of computer software and hardware, deep learning, and target detection algorithms continue to develop, and vehicle detection is gradually applied to embedded devices. Embedded devices have the advantages of small size, low price, and good stability. If accurate and real-time vehicle detection is realized on embedded devices, the development cost of the ADAS system will be significantly reduced. Firstly, according to the distribution of vehicle data and the characteristics of embedded devices, the following improvements were made to the original Solid State Drive network (SSD): (1) Using the K-Means clustering algorithm to re-set the regional candidate frames of SSD, making them more in line with the scale distribution of vehicle data and improving the detection accuracy of the model; Based on the original SSD loss function, the rejection loss is increased to improve the detection performance of overlapping vehicles. The Mobile Net V1 deep neural network is used as the feature extraction network of SSD, which can greatly reduce the amount of computation without reducing the detection accuracy, to meet the real-time requirements of running vehicle detection algorithms in embedded devices. The embedded system takes Samsung Exynos 4412 microprocessor-based on ARM architecture as the hardware platform, and transplants Linux operating system and all related drivers on this platform. In application development, QT graphical interface library and Caffe deep learning framework library are transplanted.