An Enhanced Model for Detecting Distracted Driving Behavior with Multi-Level Feature Fusion and Efficient Target Extraction

Distracted driving behavior target detection plays a central role in contributing to the improvement of public health and fire prevention. Given the limitations of existing techniques in multi-scale fusion, feature fusion efficiency, and accurate extraction of key targets in complex data environments, this paper proposes an advanced and efficient innovative algorithmic model-BEC-YOLO, which focuses on solving these challenges with a view to improving the overall detection performance. In this paper, BiFPN is firstly adopted to realize multi-level feature fusion in order to make up for the shortcomings of existing target detection algorithms in feature extraction and recognition, which effectively improves the feature fusion effect. Then, ECA is introduced to significantly improve the efficiency and accuracy of key target information extraction in complex data environments. Finally, the feature representation and fusion efficiency are further enhanced by introducing CBAM. The experimental findings demonstrate a notable 9.8% enhancement in mAP when comparing the approach in this paper to the original model. This highlights the accuracy of the BEC-YOLO model for detecting distracted driving behavior in intricate settings.