Vision-based Driver’s Seat Belt Detection

This paper focuses on seatbelt detection for assisted driving scenarios. Seatbelt detection is important to ensure driver and passenger safety. The algorithm ensures high classification accuracy against practical constraints like dynamic environments, improper in-vehicle illumination, low-quality images, and view angle variations. The paper proposes a real-time implementation of a system to detect the seatbelt. This system uses a FAST key point detector with scale invariance, good localization, and robustness to noise. It uses the BRIEF method for the robust, highly discriminative, and efficient description of key points. The feature vector optimization is done using K-means and PCA. The effectiveness of the proposed method is verified by a comparative analysis of six different feature extraction methods run through five different classifiers and the results were studied. The most effective classifier of the lot turned out to be Decision Trees.