X.W. Yang

Pages: 115-124

Abstract
The precise classification recognition of dangerous driving behaviors can effectively reduce traffic accidents and improve road safety. To address the challenges of complex and dynamic traffic environments where driving behaviors can vary significantly, an innovative method for classification recognizing for dangerous driving behaviors has been developed, utilizing an enhanced k-means clustering algorithm. Wavelet decomposition is applied to denoise the image data of driving behaviors. The refined data is then effectively categorized using the improved k-means clustering technique to discern distinct driving actions. Convolutional neural networks are employed to extract and classify critical action areas, enabling precise identification of hazardous driving behaviors. Test results demonstrate that this designed method achieves a maximum classification recognition rate of 98% for dangerous driving behaviors, with shorter recognition processing time.
Keywords: improving k-means clustering; transportation; dangerous driving behavior; classification recognition; convolutional neural network