Maritime traffic is an important part of global trade and requires advanced monitoring systems to ensure safety and efficiency. Traditional monitoring systems such as Automatic Identification System (AIS) and radar systems, while important, are limited in tracking small and non-AIS equipped vessels, especially in complex areas. This thesis addresses these shortcomings by integrating video surveillance with advanced computer vision techniques, focusing on a case study in the Port of Split, Croatia. Using high-resolution video camera and Convolutional Neural Networks (CNNs), a real-time maritime traffic counting system has been developed that improves detection capabilities beyond conventional methods. A new and diverse image dataset was created to train and evaluate CNN models for applications in the Mediterranean maritime zone. An evaluation of the counting system shows that it detects 386% more maritime traffic than AIS, with an accuracy of 98.23%. Furthermore, a novel approach using a monocular camera and the pinhole camera model is presented for distance estimation in maritime applications. By incorporating this distance estimation into the system, more accurate distance estimates are achieved, allowing the system to provide more valuable and precise data on vessel positions and movements. The enhanced detection and monitoring capabilities provided by this system highlight the effectiveness of integrating video surveillance with advanced computer vision techniques in complex maritime environments.