Abstarct: In recent years, road cameras are used in most of the roads to monitor the vehicles. Due to the increasing number of these cameras and large number of vehicles, using human resources for these cameras, makes some problems in the future. So, it is desired to use machine vision and image processing methods to estimate some parameters like vehicle speed, vehicle class, and the number of passing vehicles. Previously proposed methods have some limitations like non-automatic nature, shadow sensitivity, single-purpose, and use of small and non-real datasets. In this thesis we propose a method for speed and dimensions estimation of the vehicles. This method must meet some objectives like: fully automatic calibration, shadow resistance, multi-purpose, and provide us with a comprehensive dataset. The proposed method is conducted as follows: In the first frxames, vanishing points and the hypothetical road surface are obtained baxsed on the direction of vehicles' movement. A GMM baxsed foreground detection and a new shadow removal algorithm is applied to determine the bounding box of each vehicle and construct its 3D surrounding box. To determine the metric coefficients, some popular vehicles are recognized in the first frxames. Then, metric coefficients are calculated baxsed on the real dimensions of the vehicles (in meter) and their corresponding dimensions on the road surface (in the pixel). These steps, satisfies one of our goals, which is fully-automatic calibration. Finally, the passing vehicles are projected on the hypothetical road surface, and their precise speed and dimensions are calculated by tracking. Robustness of the algorithm against shadow leads to the accurate bounding box detection of each vehicle and reduces the errors of calibration, and speed and dimensions estimation. In the shadow removal algorithm, in addition to the accuracy, the processing time is important. So, we proposed a fast algorithm that its processing time is almost negligible. Currently, there is no standard dataset of videos including speed tags and the vehicles' images, especially in Iran. Another objective of this thesis is to provide a dataset including such information. For that, we prepared such dataset by simultaneous capturing of the roads with IP and laser camera. The calibration process needs to recognize some popular vehicles at first. Here a powerful algorithm is presented by deep networks and image registration. According to the installation of the road cameras in high altitudes, vehicle images are so small. A super-resolution algorithm is used to produce high resolution images and improve the recognition rate. The other objective of this thesis is to be multi-purpose. Most of the existing ITS algorithms are about speed estimation or vehicle type recognition. The dimensions estimation and vehicle recognition in this thesis, along with the provided dataset of the images and videos make it possible to recognize the passing vehicles by two criteria: "recognition algorithm" and "the estimated dimensions". Hence, as the secondary application of the algorithm, a method is presented to determine the type of passing vehicles, in addition to the estimation of the dimensions and speed of the vehicles. The average errors of speed and dimensions estimation are 1.2Km/h and 8%, respectively. The results show the superiority of the proposed method against previous methods in both criteria. The processing speed of the proposed method is about 2.16 FPS using MATLAB 2016 on a computer with 16GB RAM and a 2.6 GHz Core i7 processor.