In this paper, we proposed a method of parking space detection based on covariance matrix. The challenges of parking space detection come from luminance variation, shadow, and the interference of environment. For our solution, first of all, we extract Y-component from YCbCr color space, using Sobel operator for each pixel level with the brightness of the vertical gradient, and then construct the feature vector. Next, we use covariance matrix to describe a parking lot. Finally, we make parking state (free/occupied) with the distance of the covariance matrix measure characteristics. Compared with the previous methods, it has a great advantages of one-collection implement multiple identify parking Spaces, and a strong robustness.