Aiming at the illumination consistency of complex scenes in Augmented Reality (AR) system and analyzing the marker images by High-Dynamic Range (HDR) technology, an improved measurement algorithm for illumination direction based on the analysis of halo in HDR images was proposed. In order to improve the immersion and reality of virtual objects, after researching and analyzing the existing illumination recovery algorithms, a camera calibration method was proposed by utilizing the projection invariance of the quadratic curve pair. In order to get detailed light information, HDR was used to process marker image to improve accuracy. Refering to Lambert illumination model, the light information of image was analyzed to classify the shooting angle, and the improvement of traditional light source direction measuring was realized, part of the directions of the light sources outside of the photography ball reflection range was measured. The shooting 1 and shooting 2 of the single point light source were tested and analyzed. The experimental results show that this method is simple, robust, and can measure the direction of partial illumination outside the mirror ball reflection range no matter whether the marker is partially shaded or not.