Abstract: GPU real-time image rendering traditional tracking algorithm using blind separation of multi-source node scheduling and allocation method, the overall low efficiency, image rendering effect is not good. Put forward a similar target distribution field real-time graphics rendering algorithm based on the tracking measurement. Based on the natural hierarchical image, retaining the basic information of the original image, the field distribution in the object tracking algorithm, matching is performed in the image domain, target detection using gradient descent method, in the optimal position, field and the candidate distribution field similarity can obtain the maximum response distribution of the object, thus to realize the goal of field distribution similarity measure, to improve the tracking ability of real-time graphics rendering. The simulation results show that, using the algorithm of real-time graphics rendering, can improve the efficiency of rendering tracking, and the similarity is low, to overcome the multiple instance learning track and field distribution of insufficient tracking slow, easy to fall into local minimum, the superior performance of traditional method.