In order to improve the accuracy of FBG threedimensional shape sensing reconstruction, this paper proposes a temperature compensated detection method based on nonorthogonal curvature, utilized three FBG arrays to constitute sensor networks. Based on analysis of the system error components and the error causes, the influence of each component error on the 3D reconstruction accuracy is obtained with simulation model. The results show that the key issueto reduce the installation angle error can improve the 3D reconstruction accuracy. A new calibration method is proposed due to that the difference between the calibration direction and the installation direction brings the change of the measurement calibration parameters. The experimental results indicate that the proposed method can reduce the maximum error ratio of 3D reconstructed to 0.16%, which provides a reliable detection method for high precision 3D shape sensing system.