Research On Safe Driving Behavior Of Bicycle Based On Machine Learning

With the advance of green travel concept,more and more people choose bicycles as a way of travel.The bad behavior habits and complicated road conditions of the riders during the riding process can lead to accidents,thus identifying and monitoring people's riding behavior.It has very important practical significance.The goal of this paper is to collect human behavior data during bicycle riding by mobile phone sensor,data fusion of the four sensors data collected,and then use the threshold method and machine learning algorithm to design the classifier to realize five kinds of bicycle riding.Identification of behavior.The identification of riding behavior not only helps to analyze the data and characteristics of residents' riding behavior,but also forms traffic big data.It can also send early warning signals during accidents during the riding process,which can ensure the safety of travel and the monitoring and management of road traffic.The main contents of the study are as follows:(1)Data collection and pre-processing.For the various sensors on the mobile phone,in order to obtain the attitude and speed information of the riding behavior,the acceleration sensor,the gyro sensor,the magnetometer sensor and the GPS(Global Positioning System)positioning sensor on the smartphone are selected as the data source,and the mobile terminal data acquisition APP is adopted.The raw data is sent to the PC network database management system.There are various missing values and noises in the collected data.The sensor data is completely pre-processed on the PC side: Lagrange interpolation method is used to process the missing values for data integrity;window sliding mean filtering with 2s is also selected.Denoising the original signal,experiments show that the sliding mean filtering achieves a better effect.(2)The multi-sensor data fusion algorithm is studied.Aiming at the problem that the mobile phone coordinate system is inconsistent with the bicycle coordinate system in actual application,the initial relative attitude measurement method based on the combination of magnetometer and accelerometer is adopted.Experiments prove that the initial relative attitude matrix can be measured quickly and in real time.The data of the mobile phone is converted into the coordinate system of the bicycle carrier.Then,based on the quaternion-based extended Kalman filter algorithm,SINS(Straight Inertial Navigation System)/GPS attitude and trajectory data fusion,real-time update to obtain the carrier Attitude angle,velocity,bit trajectory,acceleration,angular velocity data.(3)The behavior recognition based on the combination of threshold and support vector machine is studied.In reality,the pattern recognition behavior detection algorithm has low real-time performance and high false positive rate.This paper uses two judgments to perform behavior detection through kinematic data obtained by information fusion.Firstly,according to the set threshold,the normal behavior and dangerous behavior during the riding process are preliminarily determined.The feature space is constructed according to the statistics of velocity,acceleration,angular velocity,attitude angle and trend.The splitting method is used to optimize the feature space and finally construct the most.The excellent feature space finally obtains five behavior classification models.Among them,the threshold method has a recognition rate of dangerous speed of 100%,and the support vector machine model is accurate for left/right turn,acute shift behavior,dangerous collision behavior and fall behavior.The rates reached 94.2%,90.8%,96.3%,and 95.4%,respectively.