In order to realize the indoor precise positioning of UAV under the conditions of no radio frequency positioning equipment deployment and no GPS signal

proposed an improved local optimal segmentation method. First

calculate affine transformations to equalize image grid motion statistics

and maximize the local features through the Hessian matrix; Secondly

a grid reference unit is used to detect pixel group pairs

and an optimization function is introduced to increase the weight of the grid where the feature points are located. The highest membership degree is obtained by comparing it with the neighboring grid. By combining the Progressive Sampling Consistency (PROSAC) algorithm again

feature point pairs with lower confidence are eliminated through threshold setting. Finally

the Decoupling Rotation Translation (DRT) strategy is adopted to complete the Inertial Measurement Unit (IMU) initialization pre integration and solve the spatial pose sequence. Set up flight experiments under weak lighting

complex textures

and viewpoint transformation conditions; Analysis of flight logs shows that the feature matching accuracy of multiple experiments is above 97%

and the running time is only 46%~72% of that of other combination algorithms. The indoor positioning accuracy of drones reaches 0.02 m

and the matching effect is good in a comprehensive environment. It solves the problems of long feature matching time

low accuracy

and inaccurate matching results

and the positioning is accurate

which has high application and reference value.