Environment Perception Menthod Of Intelligent Vehicle In Urban Traffic

As one of key technologies in vehicle active safety area, in the past two decades,environment perception has always been greatly concerned and rapidly developed.Many factors in urban area such as multifarious road structure, vehicle types, complexenvironmental background and frequent haze caused by environmental polution areexpected to bring huge challenges for environment perception. For these issues, thispaper condusts in-depth discussion on exploring efficient and robust environmentperception methods of intelligent vehicle to improve vehicle safety based on machinevision and vehicle communication technology.In order to improve the low visibility and poor contrast of video, we propose ahaze removal algorithm based on guide filtering method. Firstly, we simplified theatmospheric attenuation model. Then, the current concentration of haze is estimatedbased on dark channel priori theory. After that, a guide filter of brightness channelwas used to obtain current haze coverage. Videos are recovered based on the results ofestimation of haze concentration and coverage. To improve the efficiency of thealgorithm, highly time-consuming stage of haze concentration is implemented ininitialization phase, and an indicator of video definition is used to control theprocedure of subsequent module. Experimental results show that the algorithm caneffectively improve the contrast and sharpness of video while with highlycomputational efficiency. The proposed method can meet the needs of video hazeremoving.For the purpose of improving the adaptability of road model for complexenvironments, achieving a balance between efficiency and robustness in urbansemi-structured road environment, a novel road detection algorithm based onuncertain Bezier deformable template was improved. Firstly, we adaptively settwo-tier region of interest containing static and dynamic area for IMP according tocamera parameters and vehicle status and proposed Gaussian mixture directional filterfor pretreatment. Then, we construct uncertain road deformable template using Beziercurve, and convert the road recognition problem to template parameter hypothesistesting problem. We also use a modified RANSAC algorithm to solve the uncertaindeformable template parameters. At last, in order to improve the speed of solving, we proposed a level search optimization algorithm which using a combination of coarsesearch and fine search method. The test results show that the method can rapidly andaccurately extract the lane and have good immune function for typical roadsinterference in the urban road environment.Robust and reliable driver assistant system with low cost and power consumptionremains a standing challenge nowadays in vehicle active safety region. Within thispaper, we introduce here a simultaneous detection and tracking framework for robuston-board vehicle recognition system based on monocular vision technology. Thisframework takes a novel layered machine learning and particle filter approach to builda multi-vehicle detection and tracking system. It possesses strongpoint of reducedprocessing time and high accuracy thanks to the combination of pavement estimationwith vehicle identification. Concerning vehicle detection module, we have two maincontributions. Firstly, we presented a layered machine learning method that combinescoarse-search and fine-search to obtain the optimal target in whole image usingAdaBoost-based training algorithm. Moreover, we proposed a new method to estimatethe most likely pavement area on the road ahead. With regards to vehicle trackingcomponent, our main contribution is a multi-objective tracking algorithm based ontarget mode management and particle filter. The experimental results clearly showthat the proposed framework yields a robust and efficient on-board vehiclerecognition system with high precision, strong robustness and appropriatecomputational cost. The processed videos are available on a web page associated withthis paper.As type, color, size and other uncertainties are likely to cause unstable andunreliable for following vehicle detection algorithm, this paper proposed a novelcost-effective system, based on longitudinal projection in time-space domain andrelative motion analysis. The algorithm used directional filter in contrast space andthen estimated vehicle edge with probabilistic Hough transform. According totime-domain projection, following vehicle tracking trajectory can be obtained.Furthermore, this paper used optical flow to track characteristic corners, in other toseparate target vehicle from background. On this basis, the algorithm integrateddetection results of the above two methods, and then proposed an overtaking warningalgorithm based on behavior detection. The test results show that the method canextract myopia campaign following vehicle quickly and accurately in the structuralroad, and it also has a strong immunity to environmental interference and typechange. The lane departure warning methods and collaboration overtaking method withinhaze environment were studied. Firstly, since the traditional departure warningmethod does not consider the speed impacting and is difficult to give an accuratewarning time, while the TLC methods is prone to give unstable warning results, thispaper prosese a novel lane warning method using spatio-temporal warning method.On the other hand, in order to ensure vehicle safety during overtaking process withinhaze environment, collaboration overtaking method is proposed based on vehiclecommunication technology. Before the process of overtaking, utilization of therelative position, velocity, yaw-angle and other current informations of vehicle, thereminder and warning decision support is propoesd under the constant speedassumption. During the overtaking process, multivariate Gaussian distribution methodis used to create a conflict potential field and a novel overtaking risk assessmentmethods based on the estimated probability is proposed.In summary, as the core technologies of intelligent vehicle, environmentperception still faces many challenges. Against this background, in this paper,environment perception method used in urban traffic is studied. Haze removalalgorithm, lane detection and tracking algorithm, vehicle detection and trackingalgorithm, lane departure warning methods and collaboration overtaking method aregiven. The results contribute practical and heuristic significance to both theory andengineering application in related fields.