Abstract: When using depth data for translation registration, the iTOF (inderct time of flight) cameras have a problem of spatially nonuniform illumination, which can lead to a relatively large measurement depth error in the peripheral area of the plane, thereby affecting the accuracy of subsequent depth registration. In order to solve the mismatch caused by this problem, a masked frequency domain NCC (normalized cross correlation) depth registration method is proposed. Firstly, the gray image of the same frame as the depth map to be registered is masked in the nonuniform areas. Then, the frequency domain NCC method is used for the gray image registration. Finally, the translation vector obtained from registration is overlaid on the corresponding depth images to complete depth registration. This method can effectively solve the mismatch problem caused by spatially nonuniform illumination within a scaling range of 4%. The relative error of pixels in the horizontal direction can reach 1.97%, and the relative error of pixels in the vertical direction can reach 1.74%. The error follows a normal distribution, proving that this registration method has stability and reliability.

Key words: indirect time of flight principle, nonuniform illumination, normalized cross correlation, deep registration