Enhancing Road Visibility by Real-Time Rain, Haze, and Fog Detection and Removal System for Traffic Accident Prevention Using OpenCV

Nowadays, both the corporate and public research sectors are interested in autonomous vehicles. Levels of unpredictability are the reasons self-driving cars have not yet made it to the consumer market. Multiple sensors are frequently used to address this, which helps the vehicle's system become more robust. The most often used sensors are radars, lidars, and cameras, but their costs can increase quickly, making them unaffordable in some markets. Utilizing fewer but stronger sensors for visualization could help with this. This resolves the issue of the decreased view range caused by rainy weather, a specific failure mode for camera sensors. The state-of-the-art object identification with distance estimation technique, You Only Look Once (YOLOv3), is tested with the Kalman filter and discrete wavelet transform with bilateral filtering as rain, haze, and fog removal strategies. With YOLOv3, filtered movies in daytime and dusk conditions were tested, and the results reveal that the accuracy has not increased sufficiently to be valid for use in autonomous vehicles. The study field has potential and thus indicates that more object identification and distance estimate techniques be considered as future work.

Перевод пока недоступен