Accurate identification of dangerous driving behaviors on urban roads

Identifying dangerous driving behaviors on urban roads is crucial for enhancing traffic safety, streamlining traffic management, and bolstering intelligent transportation systems. To address the limitations of traditional methods, such as high relative error rates, low identification rates, and prolonged processing times, a novel method based on the Hidden Markov Model (HMM) has been developed. This approach involves collecting urban road driving behavior data using smartphones, clustering the data via spectral clustering, and refining the clustering results with Box-Cox transformation. Subsequently, driving behavior features are extracted using HMM, selected through the application of random forests, and combined with a twin support vector machine to achieve accurate identification of dangerous driving behaviors. Experimental outcomes indicate that the proposed method yields a maximum relative error rate of 4.12% for feature extraction, attains a peak identification rate of 98.75%, and operates within an identification time frame of 0.16 to 0.38 seconds.