The tomographic images reconstructed from cone beam projection data with a slice thickness larger than the nominal detector row width (namely thick image) is of practical importance in clinical CT imaging, such as neuro- and trauma- applications as well as applications for treatment planning in image guided radiation therapy. To get a balance optimization between image quality and computational efficiency, a cone beam filtered backprojection (CB-FBP) algorithm to reconstruct a thick image by tracking adaptively up-sampled cone beam projection of virtual reconstruction planes is proposed in this paper. Theoretically, a thick image is a weighted summation of a number of images with slice thickness corresponding to the nominal detector row width (namely thin image), and each thin image corresponds to a virtual reconstruction plane. To obtain the most achievable computational efficiency, the weighted summation has to be carried out in projection domain. However, it has been experimentally found that, to obtain a thick image with the reconstruction accuracy comparable to that of a thin image, the CB-FBP reconstruction algorithm has to be applied by tracking adaptively up-sampled cone beam projection data, which is the novelty of the proposed algorithm. The tracking process is carried out by making use of the cone beam projection data corresponding to the involved virtual reconstruction planes only, while the adaptive up-sampling process is implemented by interpolation along the z-direction at an adequate up-sampling rate. By using a helical body phantom, the performance of the proposed cone beam reconstruction algorithm, particularly its capability of suppressing artifacts, are experimentally evaluated and verified.