Improved Parking Space Recognition via Grassmannian Deep Stacking Network with Illumination Correction

It has become increasingly difficult to quickly locate a free parking space with the growing number of private vehicles. Many parking space management solutions have been proposed. Vision-based methods are among the approaches that have received great attention due to the widespread use of surveillance cameras in parking areas. Although promising results have been reported for vision-based methods, these methods generally suffer when good quality images are not available. The performance of vision-based methods drops under conditions like low illumination. In this paper, an approach coined as Grassmannian Deep Stacking Network with Illumination Correction (GDSN-IC) is presented. The proposed method enhances the illumination map of an image before feeding it to a Grassmannian Deep Stacking Network for parking space availability prediction. Experiments on two public datasets validate the effectiveness of the proposed approach.