Font Size: a A A

Visual Detection Of Accident-prone Driving Based On Picture Information Of Steering Wheel Area

Posted on:2018-06-21Degree:MasterType:Thesis
Country:ChinaCandidate:J H HuangFull Text:PDF
GTID:2382330566487704Subject:Vehicle engineering
Abstract/Summary:PDF Full Text Request
Accident-prone driving is an important factor resulting in road traffic accidents,which can be summarized as risk driving behaviors and poor vigilance driving state.Researches on real-time detection and timely warning technologies for accident-prone driving will be an exploratory way to improve road traffic safety.By applying the vehicle monocular vision and image processing technology,this paper conducts three typical accident-prone driving behaviors including bad hand gesture on steering wheel,fatigue driving and distraction driving.By numbers of experiment analysis and calculations,some corresponding detection algorithms and reasonable judgment methods were proposed.Finally a simple on-board system was designed.Works and achievements of this research mainly include:(1)Image differences in the wheel steering process were accumulated to help designing an adaptive locating-and-segmenting algorithm for the steering wheel image area,which was independent on the camera mounting position and photo-taking angle.Compared with the traditional methods,it just has to be sure that the steering wheel area was contained in the image acquired,and the camera needs no calibration at all.(2)Based on the differential results between the images of the driver's hand,a opisthenar's moving trajectory image extract algorithm was designed.Then a named K-step peak gray value matrix of the trajectory image sequence was designed.By looking up the matrix's minimal gray value difference path,an adaptive detection method for the initial gray value of the opisthenar was proposed.This could ensure the independence between the method and the drivers,make it not necessary to preset the driver's biological information,which is helpful to improve the adaptability and reliability of the system.(3)Within the steering wheel image area,candidate images of the driver's opisthenar were extracted based on the recognized real-time gray value information of the hand skin color.By filtering the area ratios of the candidate images against the steering wheel image,an image segment algorithm for the opisthenar part was designed.With the help of the opisthenar image segment result,a judge rule for whether the driver had put a hand on the steering wheel or not was proposed.(4)When the driver held the steering wheel with at least one hand,we proposed a feature point detection algorithm for the driver's hand web using corner detection and curvature maximum filter.Further then we could detect the driver's hand gesture on the steering wheel by judging whether the feature point existed or not.(5)In situation when the driver held the steering wheel in a correct gesture,an estimate algorithm for the steering-wheel's rotation angle was designed.By analyzing the characteristic of the angles,a method to judge whether the driver has performed a vehicle direction rectify behavior was invented.Parameters of this behavior were helpful for real-time detecting of fatigue driving and distracted driving.(6)Finally,based on OpenCV and C ++,the software of the above algorithm is realized.The hardware system is designed based on the raspberry chipset.Further realize the integration of the hardware and software.The results show that the detection accuracy of whether the driver had put a hand on the steering wheel or not is 94.3% and the recognition rate of position of driver's hand web is 92.0% and the range of estimated error of steering wheel angle is 1.6%.The algorithm has better recognition effect and reliability,also can adapt to a variety of environments.This algorithm can provide parameters for the driver's fatigue driving and distraction driving which may exist.
Keywords/Search Tags:detection of accident-prone driving, risk driving, poor vigilance driving, machine vision, image processing
PDF Full Text Request
Related items