Traditional contact-based length measurement methods for elevator compensation chains suffer from inaccuracies caused by slippage during fixed-length production. Furthermore, existing non-contact methods are unsuitable for these chains due to the low flatness, poor reflectivity, and lack of distinctive features on their plastic coating. To solve these problems, this study developed a non-contact online length measurement device for compensation chains based on eddy current sensing and a sliding-window peak detection algorithm. First, the length calculation formula was derived from the chain′s link structure, and its measurement uncertainty was analyzed to confirm the feasibility of the indirect measurement method based on counting links. Second, based on the cross-linked structure of the internal iron rings, a quasi-sinusoidal signal is formed by the variation in distance between the iron rings and the sensor probe as the rings pass an eddy current displacement sensor. The correspondence was established between the number of peaks in this signal and the actual number of chain links. Third, a one-dimensional Kalman filter was used to process the acquired signal. An online peak detection algorithm based on a sliding window was proposed, which accurately identifies the number of peaks in real-time under limited hardware resources. Finally, the measurement device was constructed. It employs a dual-cylinder clamping mechanism and a roller conveyor to reduce the interference of mechanical vibration on the acquired signal, achieving stable length measurement for various types and sizes of compensation chains. Experiments conducted on an actual production line showed that when the chain speed is between 0.5 and 1.0 m/s, the device can measure different types and sizes of chains with a maximum relative error not exceeding 5‰. This result verifies the feasibility of the method and the good stability and accuracy of the device.