Research On The Key Technology Of Driving Event Recognition Based On Android Smartphone Sensors

There is no doubt that smart phone has become a necessity in people’s everyday life.Especially the mainstream intelligent mobile operating systems such as Android andiOS enable smart phones to expand as a desktop computer. Smart phones bring accessto anyone any entertainment and information wherever and whenever possible. Fromthe time that Apple’s innovation of embedding sensors in iPhone4, mobile phones isno longer just a device with the functions of SMS message or calling, but also equipwith a stronger ability of ‘intelligent perception’. For now built-in sensors of smartphones are mainly used to develop electronic compass, motion sensing games orimproving user experience such as screen rotation when reading e-books.And along with great increase of the quantities of urban vehicles, trafficcongestion has seriously affected people’s daily life. What is more seriously, frequentserious traffic accident has caused concern of whole society. The statistical analysis ofrelevant data of domestic and foreign point out that, the main reason for the trafficaccident is from the drivers own, while the surrounding environment and vehicle witha little proportion. Meanwhile, drivers’ behavior would be relatively more secureunder the supervision environment.In this paper, sensors built-in Android smart phones are utilized to the area ofdriving recognition. Vehicle acceleration is an effective clue to reflect the currentdriving situation，such as normal acceleration and deceleration or nasty brake orurgent turning. With the power of ‘perception’ no matter wherever and whenever,smart phones placed with a particular orientation inside a vehicle can collect host’sreal-time acceleration which made it possible to analyze the motion state of thevehicle.There are two main algorithms for patter recognition. One is Dynamic TimingWarping (DTW) which is widely applied in isolated-word speech recognition; anotheris Symbolic Aggregate Approximate (SAX) which is used to measure the similarity oftwo time series in data mining. Compared with DTW, SAX has an advantage of timecomplexity. However, SAX is easily to lose extreme values in the process ofapplication of PAA to data segment and dimension reduction. According to the drawbacks of the algorithms of DTW and SAX, this paperexploits the Extended SAX algorithm on the basis of SAX algorithm to improve theaccuracy of identification of driving event. The Extended SAX algorithm retains theextreme values of each segment of PAA division then expresses all the segments as asymbolic sequence.In addition, in order to eliminate the influence of vehicle’s own gravity, in thispaper applies a rotation matrix which is obtained through the acceleration sensor andmagnetic sensor to calibrate. The rotation matrix makes it possible to turn the devicecoordinate system to the world coordinate system (WCS).By the testing experiment, it proves that the application of E-SAX algorithm caneffectively identify the driving event. In the final, analyze the multiple factors thatmay influence recognition rate and point out the improvement ideas in the future.