Object-shape Recognition Based On Non-crosstalk Ultrasonic Ranging

With the tasks and environments getting more and more complicated, mobilerobots are required to be more intelligent and automatic for the purpose ofestablishing the map of environments and performing a variety of tasks. Obstaclerecognition is the premise of environment modeling. Therefore, it is necessary tostudy obstacle-shape recognition technology. Because of its advantages, ultrasonictransducers have been widely used in ranging systems for object recognition.In a multi-channel ultrasonic ranging system, crosstalk and the contradictionbetween effective ranging distance and ranging resolution are two main reasons thataffect the accuracy of the ultrasonic ranging system. To eliminate crosstalk, theexisting shape-recognition methods usually adopt time-sharing strategy, which needsmore time to get distances. To get high range resolution, single pulse wave is usedwidely, which has a negative effect on the ranging distance.To solve these problems, the idea of exciting ultrasonic transducers withpseudorandom sequences is proposed in this thesis, and the correlation-processingtechnique at the receiving end is adopted to measure the distance. On this condition,the crosstalk phenomenon can be eliminated and the real-time performance can beimproved as well, and a longer effective ranging distance as well as the resolution canbe obtained at the same time.First of all, both the transmission model and the characteristic of thetransmission/reflection of ultrasonic are introduced. How these characteristicsinfluence the ranging and shape recognition is analyzed. The design and theimprovement of ultrasonic-transducer array are explained.Secondly, the application of neural network to pattern identification is discussed,and a neural network structure is designed for this system. Based on simulations andexperiments, a lot of data have been collected and analyzed. A training program isgiven by neural-network toolbox of Matlab, and the weights and biases of the neuralnetwork are saved for recognition.Lastly, on the foundation of the non-crosstalk ultrasonic ranging system builtpreviously, the object-shape recognition is discussed. The results show that the simpleshapes, including plane, corner and edge, can be identified.