The camera response function (CRF) establishes a numerical mapping between focal plane radiance, camera imaging parameters, and the intensity of output images. It plays a significant role in areas such as high dynamic range imaging and image processing. To establish an accurate response model for time delay integration (TDI) complementary metal-oxidesemiconductor (CMOS) imaging systems, this paper proposes a radiometric calibration method for TDI CMOS imaging systems based on complex real-world scene images. The study begins by conducting an extensive analysis of the data link within the TDI CMOS imaging system, which serves as the foundation for establishing its a priori theoretical response model. Subsequently, the problem of solving the CRF model is transformed into an overdetermined equation established through the least square method. The optimal solution for this equation is obtained by singular value decomposition (SVD), which leads to the derivation of a three-dimensional response function for the imaging system. Finally, under consistent optical radiation conditions, radiation calibration experiments are performed on various targets using a self-developed TDI CMOS imaging system. The CRF is obtained based on the captured experimental image data. Furthermore, this paper's approach is compared with the widely adopted linear fitting method commonly used within the respective field. The experimental results show that the visually perceived quality, structural similarity, mean grayscale, mean gradient, entropy, and standard deviation of images synthesized using the CRF proposed in this paper are closer to those of actual captured images. The proposed method demonstrates higher accuracy and can provide a reliable basis for applications such as radiation response calibration of on-orbit spaceborne payloads, selection of imaging parameters, and multi-exposure fusion of remote sensing images.