More and more security vulnerabilities are found in smart cars for an increasing electronic control units (ECUs) and external communication interfaces. Intrusion detection system (IDS) has become the focus research as an initiative protection measure, yet the key issues are not well resolved, such as detection accuracy and computation overhead. According to the vulnerability analysis and the difference features of abnormal data under different attacks, the in-vehicle network attack experimental platform and the abnormal behavior characteristic library are built. Then, the low complexity learning algorithm is used to detect the simple abnormal features. Meanwhile, the further complexity detection method are proved to achieve accurate detection of abnormal data for the abnormal behavior of false data injecting and semantic tampering. Our research work makes it easy to understand intrusion detection system for in-vehicle networks and benefits the implementation of security protection in automotive industry.

In-vehicle networks;abnormal data;intrusion detection;attack scenarios