Taiwan has many vehicles and as a result many traffic problems. In particular, during the Spring Festival and the holidays, Hsuehshan Tunnel (HST) between Yilan and Taipei is always a traffic jam. To solve this problem, intelligent transportation system (ITS) is necessary, and accurate vehicle detection (VD) is the first stage for ITS. In order to detect vehicles in HST, three training methods based on single shot multibox detector (SSD) are presented to detect four categories of vehicle in the Tunnel. The experimental results demonstrated that the presented three training methods, which only used 1000 training frames, can detect and categorize more vehicles than the pre-trained SSD model which used a large training dataset. Specifically, the SSD trained by our collected data set and data augmentation has the highest detection rates for sedan, van, bus, and truck - 93.6%, 90.9%, 100%, and 100%, respectively.