Object Detection in Foggy Conditions by Fusion of Saliency Map and YOLO

Abstract

Under foggy conditions, visibility decreases and causes many problems. Less visibility due to foggy conditions while driving increases the risk of road accidents. It is important to detect and recognize the nearby objects under such conditions and predict the distance of collision. There is a need to devise an object detection mechanism during foggy conditions. The paper proposes a solution to this problem by proposing a VESY(Visibility Enhancement Saliency YOLO) sensor which uses an algorithm that fuses the saliency map of the foggy image frame with the output generated from object detection algorithm YOLO (You Only Look Once). The image is sensed using image sensors in a stereo camera which are activated using a fog sensor and a depth map is generated to calculate the distance of collision. Dehaze algorithm is applied to improve the quality of the image frame whose Saliency image is generated on the basis of region covariance matrix. YOLO algorithm is also implemented on the improved quality image. The proposed fusion algorithm gives the bounding boxes of the union of the objects detected by Saliency Map and YOLO Algorithm thus proving to be a viable solution for real-time applications.