Weather and Light Level Classification for Autonomous Driving: Dataset, Baseline and Active Learning

Autonomous driving is rapidly advancing, and Level 2 functions are becoming a\nstandard feature. One of the foremost outstanding hurdles is to obtain robust\nvisual perception in harsh weather and low light conditions where accuracy\ndegradation is severe. It is critical to have a weather classification model to\ndecrease visual perception confidence during these scenarios. Thus, we have\nbuilt a new dataset for weather (fog, rain, and snow) classification and light\nlevel (bright, moderate, and low) classification. Furthermore, we provide\nstreet type (asphalt, grass, and cobblestone) classification, leading to 9\nlabels. Each image has three labels corresponding to weather, light level, and\nstreet type. We recorded the data utilizing an industrial front camera of RCCC\n(red/clear) format with a resolution of $1024\\times1084$. We collected 15k\nvideo sequences and sampled 60k images. We implement an active learning\nframework to reduce the dataset's redundancy and find the optimal set of frames\nfor training a model. We distilled the 60k images further to 1.1k images, which\nwill be shared publicly after privacy anonymization. There is no public dataset\nfor weather and light level classification focused on autonomous driving to the\nbest of our knowledge. The baseline ResNet18 network used for weather\nclassification achieves state-of-the-art results in two non-automotive weather\nclassification public datasets but significantly lower accuracy on our proposed\ndataset, demonstrating it is not saturated and needs further research.\n