An image shadow in an autonomous vehicle often causes failures in image segmentation and object tracking and in recognition algorithms. In this paper, a shadow detection method based on a support vector machine (SVM) is proposed. Firstly, an RGB image was converted to LAB color space, and a shadow detection model based on an SVM was obtained by training the image with a shadow. Then, the image was divided into a shadow region, a shadow boundary, and a light region. Moreover, the light intensity in the shadow region was adjusted by eliminating the pixel difference between the shadow region and the light region. Meanwhile, the image gradient was established within the shadow boundary, and the boundary shadow was replaced by smooth interpolation to achieve a smooth transition from the light region to the shadow region. Finally, a clear image without a shadow was recovered using wavelet gradient data. Experimental results show that this method can detect the shadow region in an image and reproduce the image without the shadow effectively.