Driving posture recognition by convolutional neural networks

Driver fatigue and inattention have long been recognized as the main contributing factors in traffic accidents. Development of intelligent driver assistance systems with embeded functionality of driver vigilance monitoring is therefore an urgent and challenging task. This paper presents a novel system which applies convolutional neural network to automatically learn and predict four driving postures. The main idea is to monitor driver hand position with discriminative information extracted to predict safe/unsafe driving posture. In comparison to previous approaches, convolutional neural networks (CNN) can automatically learn discriminative features directly from raw images. In our works, a CNN model was first pre-trained by an unsupervised feature learning called using sparse filtering, and subsequently fine-tuned with four classes of labeled data. The Approach was verified using the Southeast University Driving-Posture Dataset, which comprised of video clips covering four driving postures, including normal driving, responding to a cell phone call, eating and smoking. Compared to other popular approaches with different image descriptor and classification, our method achieves the best performance with a overall accuracy of 99.78%.