The state of a road encompasses various factors, including the presence of vehicles and pedestrians. On-the-road surveillance cameras are crucial in tracking information regarding pedestrians, moving vehicles, and objects in the road vicinity. With the increasing adoption of deep learning models in vehicle monitoring systems, there is a focus on identifying and classifying conditions relevant to driving comfort and road safety. This study addresses the issue of an imbalanced dataset. However, the scarcity of publicly available datasets and efficient detection techniques poses challenges in accurately classifying road situations. In this paper, we propose a long-term recurrent convolutional network (LRCN) comprised of two groups integrated with convolutional neural networks (CNN) and long short-term memory (LSTM) networks with multiple layers of LSTM blocks for road situation detection in videos. Our model for road situation classification achieves an accuracy of 71% when using the original dataset captured from the camera. However, employing data augmentation to balance the dataset resulted in an increased accuracy of 90%.