Lane Detection Design And Implementation Based On Android Mobile Phone Sensors

With the continuous growth of the number of cars and the development of intelligent transportation,traffic safety has always been a problem of great concern to people.Driving in incorrect lanes and unsafe lane changes is a major factor in traffic accidents.If drivers can be timely alerted of driving lanes inaccuracy,many traffic accidents can be avoided.At present,the research on lane recognition is mainly based on vision-based lane detection technology,which is one of key asseistant technologies for driving safety of automobile.However,the technology is greatly influenced by external factors.For example,the fuzzy road marking and foggy rain and snow weather will give a great impact on the accuracy of lane detection.Based on the above background,this thesis presents a lane detection scheme based on android smart phone sensor and moreover designs and implements prototype system.The contributions of this thesis are the following threefold:1.First,we propose a scheme to detect the target lane which the vehicle enters after turning at the intersection,based on the idea that different target lanes correspond to different track radius.Specifically,the turning process of the vehicle is regarded as a periodic circular movement.The radius changes of the circle corresponding to different driving trajectories are different.Using the sensing data gathered by three-dimensional acceleration sensor and the gyro sensor equipped on the smart phone,the turning radius of the vehicle is infered to obtain the radius of the circle over time.The feasibility of the proposed scheme is verified through experiments.The thresholds for distinguishing different target lanes are set according to the obtained radius data.The accuracy of the proposed scheme is 70%.2.Machine learning based classification algorithms have been widely used in human behavior recognition.Here,classification algorithm is utilized to detect various vehicle behaviors.The typical seven vehicle behaviors that affect the lane location are identified,and the straight-ahead and stationary behaviors are also considered.The data acquisition program is built and the data of the nine behaviors are collected.Then a series of data preprocessing,feature extraction and training data are selected.In the feature extraction,the commonly used time-domain and frequency-domain features are summarized.Through experimental analysis on Weka,the appropriate features areselected.In the process of vehicle behavior recognition,the common classification algorithms used for behavior recognition and the advantages and disadvantages of these classification algorithms are summarized.The accuracy of these classification algorithms in vehicle behavior recognition is analyzed experimentally and it is found that the random forest classification algorithm has the highest accuracy of 84.13%.3.Combining the above two parts,a lane-level positioning system is designed and implemented which can update the position of the vehicle lane in real time by detecting the behavior of the vehicle when the initial lane of the vehicle is known.The system realizes the sensor data collection,upload and the final positioning lanes display on the APP side,and implements the energy consumption work such as feature extraction and algorithm model identification on the server side.