CAR VISION IN EXTREME WEATHER CONDITIONS

Abstract

Self-driving vehicles must be able to properly recognize traffic obstacles in real time in order to be useful in preventing accidents. Sensors like cameras are commonly employed to help autonomous cars track and monitor their surroundings. In low-visibility circumstances, such as snow, rain, or fog, these sensors, on the other hand, have a hard time adjusting. As a result, they can’t read or detect signs, pedestrians, or other cars. In Africa and the Middle East, fog and sandstorm situations are very dangerous restrictions on cameras and sensors as these situations reduce visibility, and affects driver’s safety. Thus, in this thesis, we use state-of-the-art deep learning algorithms. First, we use “Transformers” algorithm to enhance the image under these conditions. Then, we use “YOLOX” (You-Only-Look-Once) algorithm for detection in order to compare between number of detected objects in the enhanced images and their original form. As a result of the enhancement, the model ended up successfully detecting more objects in the enhanced image rather than noisy one.