Driver's Seat Belt Detection Using CNN-SVM: A Hybrid Approach

Seat belt detection is essential in transportation systems to reduce accidents. A seat belt can prevent the driver from fatal injuries. Because the seat belt is a major safety feature inside the vehicle the driver did not give much importance to the seat belt as a safety feature. But seat belts can save the lives of drivers and passengers in more than 40-50% cases of vehicle accidents. We are using a lot of new technology such as sensor-based but they are not very effective in increasing the number of vehicle users with a seat belt. Current technology for generating remainder for seat belts has many shortcomings, It can easily be fooled by using a Seat belt warning stopper or we can buckle the seat belt under the arm. We propose a new technique to detect through the image of the driver and this technique is called a hybrid CNN-SVM classifier for a seat belt. In this technique, the image has a buckled seat belt in the front of the driver's position inside the vehicle. Convolution Neural Network is basically a deep Neural Network. SVM is used as a binary classifier to increase the model's accuracy. In this case, we use CNN to the extraction of features and classify images SVM (Support Vector Machine) is used for the classification of images. We obtained 98.50% accuracy which is more than other traditional or pre-trained models.