Abstarct: The transportation industry is one of the most active industries in the world and road safety is one of the most important areas of research and development in this industry. Lane detection system and obstacle detection system are among the safety equipment used in smart vehicles to navigate the vehicle between the road lines and prevent vehicle deviation and accidents in dealing with obstacles. Video images contain very valuable information about the environment around the car and play a key role in detecting road lines and obstacles. Due to the fact that we are faced with a natural environment in recognizing road lines and obstacles in video images, there are several factors such as different weather conditions, the shadow of buildings and trees along the road, the shadow of cars, the state of sunlight at different hours of the day and the light of headlights of oncoming cars at night, make it difficult to detect road lines and obstacles. On the other hand, due to the large volume of computations for processing image information and also the need for real-time detection of road lines and obstacles, reducing computational complexity and increasing computing power should be considered for the proposed system. In this thesis, the aim is to identify road lines and obstacles in front of the vehicle using video images taken from the road, and to determine the driving conditions on the road. For this purpose, we have used two different proposed methods in this study. First, we used a classical approach baxsed on basic image processing algorithms, and using edge detection and Hough transform algorithm, to identify road lines with an unsupervised approach. One of the innovations in this method, in addition to being real-time, which has been done as the main goal in the executive method, is to perform a dynamic masking to identify the area of interest in the image. In the second proposed method, which is baxsed on a deep learning approach, we have identified a line using a Semi-Supervised Generative Adversarial Network, and then using a Histogram Oriented Gradient descxriptor, we identify the obstacles in front of the car. Among the important points in the second method is the use of a deep Semi-Supervised Generative Adversarial Network in the segmentation process, as well as its lower computational complexity compared to known deep network architectures such as ResNet50, AlexNet and VGG19. Also, in the obstacle detection algorithm, by applying the dynamic masking policy, we have achieved an faster execution speed in detecting oncoming vehicles. In fact, the masking used is a very simple idea to increase the processing speed of the algorithm. The results of applying the first proposed method indicate the accuracy of 96.78% lane detection in real time on the IROADS databaxse used in different environmental conditions. Also, the results of using the second proposed method for detecting road lines have a pixel accuracy of 90.78%, which indicates that in addition to the relatively good percentage that the implementation method has over the supervised approaches, in terms of computational complexity in the number of training parameters used, with 21 million training parameters, there has been a significant improvement over the aforementioned deep networks. Finally, it should be noted that the output of the obstacle detection algorithm in front of the car, according to the quality criterion defined for its operational accuracy, has an accuracy of 92.3% in obstacle detection.