Data feature extraction method of intelligent vehicle road system test scene based on autoencoder

Abstract

The intelligent vehicle road system can effectively improve the safety and traffic efficiency of road traffic. Fully comprehensive testing is an important prerequisite to ensure its large-scale application. How to find the limits and typical traffic scenarios is the key to carrying out this type of testing. This paper responds to the call of the national transportation power country policy and uses autoencoder to extract the typical characteristics of driving data. Aiming at the problem of driving event data with too high dimensionality and too many feature parameters for direct clustering, this paper uses ordinary auto-encoding, denoising auto-encoding and variational auto-encoding networks to train and test the driving event data. Use Tensorflow's weight extraction method to extract the optimal weights after training, and use these weights to calculate and extract driving event data features. Finally, the superiority of the denoising self-encoding network is demonstrated through data comparison. Therefore, the denoising self-encoding network can be used to extract the characteristics of driving event data.