Research On Optimized Selection Of LED Feature Points Coordinate For Pose Measurement

To achieve monocular vision pose measurement,we install multiple LEDs as the identification points on the surface of the object.According to monocular pose measurement principle,the geometric constraint of identification points is the condition to determine whether the iterative target equation converges.Therefore,the space position error of identification points will directly affect pose measurement accuracy.It is necessary to measure the coordinates of LED identification points.In monocular pose measurement process,the angles between the optical camera axis and the object vary with attitude.Considering the environmental noise and extraction algorithm and camera factors,error of image point extraction cannot be ignored.A monocular measurement system with multi-angle is developed,which could combine pose measurement and coordinate measurement and reduce the effect of image center extraction error.In view of this situation,the coordinate optimization method based on error propagation and neural network are studied.1.To learn the theory of monocular vision measurement,which includes the pinhole imaging model and the pose calculation method.The key technology of monocular multi-angle space coordinates measurement is studied,and its core content is multi-angle rotation model and non-orthogonal three-dimensional turntable motion model.The experiment system of monocular vision pose measurement is built,and the system is calibrated firstly.Monocular multi-angle LED feature points coordinate measurement are carried out.2.A monocular pose measurement error transferring model is developed to deduce the error propagation formulas from image point error to final output pose error.Based on the prediction and estimation of output pose error,feature points coordinate is selected.Pose measurement experiments are carried out,and it is proved that the coordinate optimization method based on error propagation can effectively improve the accuracy of monocular vision pose measurement.3.The optimization method of identification point coordinates based on neural network is studied.I go deep into the theory of BP neural network and extreme learning machine(ELM)method to analyze the advantages and disadvantages of neural network and extreme learning machine.Pose measurement experiments are carried out,and it is proved that the coordinate optimization method based on extreme learning machine can effectively improve the accuracy of monocular vision pose measurement.