Vehicle Speed Anomaly Detection Based on MLP AutoEncoders

Intelligent Transportation Systems (ITS) play an important role in modern urban traffic management, where monitoring vehicle speed is essential for reducing traffic violations and analyzing accidents. Traditional anomaly detection methods for vehicle speeds depend largely on labeled data and often falter when dealing with Complex problems. In response, we propose a novel anomaly detection approach using a Multi-Layer Perceptron AutoEncoder (MLP AutoEncoder). This technique employs unsupervised learning to identify deviations in vehicle speed without extensive labeled data. Our research showcases the application of an autoencoder for encoding speed time-series data across various road segments, facilitating data compression and feature extraction. We also developed an anomaly detection algorithm that leverages this encoded data to accurately identify abnormal speed behaviors on different roadways. Comparative experiments with technologies like Transformers, LSTM, and TCN affirm the superior efficacy of our model in detecting anomalies.