In order to extract detailed information about the environment, a mass of high dimensional nonlinear data information were collected by wireless multimedia sensor network nodes. The analysis and visualization of sensory data is a very important aspect of sensor network application system. In this paper, we propose Locally Linear Approximating (LLA) algorithm. It reaches the aim of locally linear approximating through adopting the way of direct gradients estimation, thus realizes the dimensionality reduction of the high-dimensional nonlinear data. The presented algorithm is proven to satisfy its goals through the computer simulations, and can provide a basis for the complex nonlinear sensory data classification and retrieval.