NETTED: Non-Standard CAN ID Extraction System Based on Deep Neural Networks

With increased interest in smart driving technology, interest in understanding the current state of the vehicle using CAN data has also increased. Standardized CAN information such as vehicle speed and engine RPM information can be easily acquired and used. However, most of the CAN data related to the driving information such as the angle of the steering wheel is not standardized. To solve this problem, this paper proposes and implements a system for extracting nonstandard CAN IDs based on deep neural networks. The system can estimate the current state of the steering wheel and brake pedal by learning data collected from GPS and IMU. Then, the appropriate CAN ID and byte position are extracted based on the estimated current state information. Experimental results show that the estimation accuracy of the steering wheel and brake pedal are 87.63% and 94.81%, respectively. Also, it is confirmed that CAN data area is reduced by about 90% through the pre-filtering process. Finally, the proposed system extracts two desired CAN IDs.