In indoor positioning for blind people

pedestrian dead reckoning(PDR)is faced with the problems of insufficient modeling of motion states and cumulative error.To address these problems

a motionstate constraints(MSC)method

named MSC-PDR

is proposed.By constructing a dynamic temporal model based on inertial data

the accuracy of motion state recognition is improved

thereby constraining the PDR process.In step detection

combined multi-threshold and time-interval constraints are adopted to suppress abnormal gait interference.In step length estimation

linear and nonlinear models are adaptively selected according to motion states.In heading estimation

weights are dynamically adjusted based on real-time motion conditions to reduce cumulative turn-induced drift.The experimental results show that the recognition accuracy is 83.36%

and the F1 score is 85.16%

which is over 6% higher than that of existing methods.When the recognition results are applied to positioning

the mean absolute error(MAE)，root mean squared error(RMSE)and standard deviation(STD)of MSC-PDR are 0.178

0.192

0.086

respectively

which are lower than those of existing methods.In four scenarios

the average cumulative error over 50 steps is 34m

and the average singlestep error is 0.17m

indicating low positioning error