For the wireless sensor network based on fingerprint indoor positioning, an indoor positioning algorithm based on kernel-based orthogonal projection to latent structures (K-OPLS) was proposed. O-PLS was a universal linear multivariate data modeling algorithm, which can remove the irrelevant components in the input and output, that was, eliminating the variation components from the description variable (input) that were orthogonal to the response variable (output). Under the model framework of O-PLS, the ‘kernel trick’ was applied by the K-OPLS algorithm to map the description variable to the high-dimensional feature space, and the nonlinear relation between the description and response variables was given, so as to achieve the calculation of the predictive component of the model and the response-orthogonal component. Therefore, the essence of K-OPLS algorithm was to integrate the kernel partial least square modeling and orthogonal signal correction preprocessing method, which can effectively improve the performance of the model and enhance the interpretation of the model to some extent. Based on RSSI fingerprint information, the K-OPLS algorithm can achieve the indoor localization tracking for WSNs by building the nonlinear mapping relation between anchor node and non-anchor node in the reference position. The proposed algorithm was applied to different examples in the simulation and physical environment. Under the same conditions, it was also compared with other kernel-based learning algorithm, such as kernel ridge regression (KRR), kernel extreme learning machine (KELM), kernel signal to noise ratio (KSNR), kernel partial least squares (KPLS), and kernel adaptive filtering etc. In the simulation experiment, the tracking error of the WK-OPLS algorithm based on wavelet kernel was 0.2326m and 1.3205m, respectively, in no-noise and noisy environments. In physical experiments, the tracking error of the algorithm based on wavelet kernel was 0.2493m. The experimental results showed that the proposed algorithm not only improved the positioning accuracy effectively, but also had a certain ability to remove noise.