Development of the Intelligent Object Detection System on the Road for Self-driving Cars in Low Visibility Conditions

Self-driving cars is rapidly developing area in modern world. Many companies, such as Tesla, are using computer vision technologies to automate driving process by computers using a lot of cameras, integrated in car bodies. One of the problems is weather. We are solving the following task: One day a self-driving car was driving and suddenly it started to rain. The main camera has lost image clarity, except random square area. We need to find out, either a man or a car is in this square area. It is guaranteed that there is a car or a man there. Classes are balanced. For train dataset we have images, positions of square areas with their size and labels. We do not have test dataset, but it is known, that it has only images. So, we need to create algorithm to locate areas, where image is not blurred and only then solve classification task. Finally, model scored using cross-validation.