In response to the issue of precise altitude measurement in complex terrains where existing proximity sensors struggle to adapt, a novel imaging-based adaptive ground altitude estimation method in the image domain is proposed. This method applies the principle of Doppler sharpening imaging to the forward-looking view scene, taking advantage of the high resolution in the terahertz frequency band and the short synthetic aperture length. It utilizes terahertz forward-looking imaging for terrain perception, transforming the one-dimensional distance measurement of traditional proximity sensors into two-dimensional imaging measurement, thus achieving precise perception of the terrain and topography directly below the motion trajectory. After obtaining the ground target image, the altitude estimation is realized by fitting image domain features using the characteristics of the ground image. Theoretical analysis and simulation verification were conducted on the parameter design and imaging algorithm of the forward-looking Doppler sharpening imaging. A terahertz proximity detection system operating at 220 GHz frequency band was developed and unmanned aerial vehicle-mounted tests were carried out. The tests show that this method can effectively perceive the ground environment, eliminate interference from corner reflectors, trees, and other targets on ground altitude measurement, and achieve an altitude measurement accuracy of 0.5 m. This demonstrates the feasibility of the terahertz imaging- based ground altitude estimation method and lays a foundation for improving the adaptability and perception capability of proximity sensors to complex terrains based on image domain features.