Driver’s Face Detection in Poor Illumination for ADAS Applications

The detection of the driver's face is an important part of the active safety system for advanced driver assistance systems (ADAS) and this task becomes challenging and complex when normal visual cameras are used in poor illumination. This paper incorporates the adaptive attenuation quantification retinex (AAQR) method to improve the details of low and poorly illuminated images and videos with the Viola-Jones face detection algorithm. This paper also makes a comparative study of the implementation of the proposed solution and analyzes its performance on three different datasets: EBDD, 3MDAD IR2, and our self-generated validation dataset. The results have shown that when AAQR is combined with Viola-Jones, the face detection has been improved in the range of 11.18% to 24.21% for different illumination categories of EBDD, 16.71% for 3MDAD IR2 (night), and 46.79% for validation datasets.