Ultrasonic Ranging Signal Optimization And Obstacle Recognition

Ultrasonic ranging system has been widely used in a robot or unmanned aerial vehicle as an indispensable sensor components with non-contact measurement,cheap,simple principle,not affected by light.However ultrasonic sensor is vulnerable to interference from other clutter.It is difficult to find the exact zero echo signal.Threshold methodare commonly used to measure the transit time which leads to the low accuracy.Information which the robot collects from the sensors are demandedhigher and higher accuracyin the complex environment.The robot must be very familiar with the surrounding environment toplanthe optimal path and provide better service to humans,thereforeit is importantfor the robot to distinguish the obstacles surrounding.At present,the sensors which include laser,vision and ultrasound are commonly used in robotics.The price of laser is expensive,although it has high precision.Generally,Vision can collect amount of information which requires lots of calculation and higher hardware,but it is affected by the light seriously.With its own advantages,ultrasound has become an indispensable sensor for robots.Finite nonlinear recursion filter with Blackman window function which the amount of calculation is small and easy to implement has a better filtering effect.The band-pass filter which is carried out through the Fourier transform of the original signal to proposed 40 k Hz demands the large amount of calculation.L approximation order polynomial from echo signal is proposed to reconstruct the echo envelope function when it can accurately find the echo signal of zero.Six ultrasonic ranging facilities are designed to identify the shape of obstacles around the robot.Data samples are from the ultrasonic ranging equipment which meets the obstacles.The sample points are consist of 160×6 matrix which are from four obstacles information.Each obstacle contains the information of 40×6 matrix.The information 20×6 matrix of each obstacle is used to the training set to train the Support Vector Machine model.The other information is used to the prediction set.Then the model is used to predict the accuracy of the obstacles by the prediction set.Through the cross validation of the penalty factor C and the kernel parameter g of the SVM model,the accurate resolution can reach 97.5% when the optimal parameter C is equal to 2.83 and g is 0.5.