In the context of increasingly complex urban traffic, the deployment of automatic systems to detect and track vehicles participating in traffic from traffic surveillance videos is becoming urgent. In recent years, deep learning methods have made many achievements in detecting complex objects. However, there are still challenges that current systems need to solve in real-life situations such as obscured vehicles, background noise, etc. To solve these problems, this paper proposes a solution combining the YOLOv11 model and the DeepSORT algorithm to improve the accuracy of detecting and tracking vehicles from traffic videos. At the same time, the combination with the DeepSORT algorithm allows tracking multiple objects simultaneously. The Kalman filter is used to accurately estimate the motion state of vehicles, helping to minimize common errors in complex situations. Experimental results show that the proposed system has effectively solved the problem of obscured vehicles, thereby improving the detection accuracy and reducing the rate of missed objects. The research results open up a promising direction in building automatic traffic monitoring systems that can operate stably in diverse real-life conditions.