Wireless Sensor Networks (WSN) provide an economical, convenient alternative solution to existing traffic monitoring systems, such as video recognition and inductive loops. However, there are many technical challenges in on-road real-time traffic information collection with WSN, e.g. accurate vehicle detection in low Signal-Noise-Ratio (SNR) conditions and reliable vehicle speed calculation. In this paper we propose a systematic solution to on-road real-time traffic monitoring with a magnetic sensor network. We propose a Similarity Based Vehicle Detection (SBVD) algorithm to detect vehicles in low SNR conditions by calculating the similarity between on-road signals and a referential signal. We propose a Collaborative Speed Calculation (CSC) mechanism to calculate vehicle speed reliably by redundant nodes and accurate reports of vehicle appearance. The CSC mechanism can calculate vehicle speeds accurately and trustworthily. We demonstrate through simulations and experiments that our proposed solution works effectively in noisy on-road environments. The vehicle detection rate is above 90% while no false alarm occurs and the error rate of speed calculation is under 10%.