The railway point is one of the most critical systems in railway, which enables trains to be guided from one track to another. Failures of railway point systems often lead to service delays and hazardous situations. With the ever-increasing railway operation, the stable and safe operation of railway point systems becomes more and more important. The electrohydraulic railway point system is widely adopted due to its high efficiency and long service life. However, its operation highly relies on a sufficient volume of oil, which usually leaks gradually due to many practical issues, e.g., the wear of joint rings and the aging and corrosion of oil pipes. The gradual oil leakage often causes failures of electrohydraulic railway point systems, so it is important to develop the effective and reliable technologies for the early detection of slow oil leakage. To achieve this goal, the static oil level is used to construct the monitoring statistic since it is a direct indicator of the oil volume and more suitable for railway point systems than the hydraulic pressure due to the discrete demand for operations and the short duration of every operation. However, it is very challenging to utilize the oil level directly for oil leakage detection due to the large variability caused by the environmental temperature. Therefore, in this article, we propose a penalized convolution-based model to capture the relationship between the temperature and the oil level, whose parameters are estimated by using least squares with a smooth penalty. With the trained model, we obtain the residuals by removing the effect of temperature from the oil-level signal. Then, the CUSUM chart is employed to monitor the residual signal since it is sensitive to the small mean shift. Finally, the proposed method is evaluated through a case study by using in situ data collected from a real point system in Changsha Railway Station.