Enhancing distracted driver detection with human body activity recognition using deep learning

Abstract

Deep learning has gained traction due to its supremacy in terms of accuracy and ability to automatically learn features from input data. The literature proposes various approaches to detect distracted drivers. The performance of these algorithms is usually limited to image datasets with similar distribution as the training dataset, preventing the successful translation of the algorithm to real-world application. This paper proposes a robust distracted driver detection approach based on recognising distinctive activities of human body parts involved when a driver is operating a vehicle. Experimental results suggest that the proposed method outperforms current deep learning algorithms for distracted driver detection and significantly improves cross-dataset performance. A classification accuracy improvement of 15% was observed. Most importantly, a significant overall balanced (F1score) performance improvement of 23% was observed.