Research On Lane Information Perception And Incremental Modeling Technology For The Application Of Intelligent Vehicles

The environmental perception and environmental information modeling are important technologies in intelligent vehicle field.The perceptual system,as the basis for the realization of functions of intelligent vehicles,is to achieve the positioning of the vehicle and the recognition of the surrounding environment by the sensor;the environmental information modelling is to establish a model applicable to decision control of intelligent vehicles by perception information fusion.An in-depth research on lane information perception of intelligent vehicles and lane information modeling technology based on incremental maps is made in this paper.The lane-level information,including vehicle positioning information,lane line information and traffic sign information,is obtained through visual perception and positioning perception technologies,the incremental digital map technology is applied for lane-level information fusion and modeling,and the timeliness and accuracy of digital maps are improved by incremental modeling technology.The main research work includes the following four parts:(1)The positioning perception algorithm based on multi-model interaction.The vehicle navigation and positioning information perception algorithm based on multi-model interaction is proposed in response to low accuracy of low-cost in-vehicle GPS positioning.In consideration of kinematic characteristics,dynamic characteristics and limiting acceleration braking capability of vehicles,the longitudinal and lateral steady-state steering motion model describing the conventional constant-velocity motion of vehicles and the adaptive variable acceleration model describing variable-velocity high mobility longitudinal motion of acceleration and deceleration states of vehicles are established respectively,the full coverage predictive description of areas that a vehicle may reach in the period ahead is realized through the state estimation algorithm such as model interaction,filtering and probability updating,and a direction-based data association filter which can better predict the driving state of the vehicle is proposed to achieve the high-precision track plotting based on low-cost GPS and INS.(2)The lane-line perception algorithm based on gradient features.When a vehicle is in a driving state,it will be affected by traffic environment,weather conditions,light changes and complex motion states of the vehicle,which will affect the performance of the visual perceptual system.In order to reduce the influence of the above factors,in this paper,an FPGA-based real-time visual perception hardware system is developed,which can process 720 P image data at 25 FPS to meet the real-time requirements of the environment perceptual system;in the light of the features of embedded system of FPGA,the image preprocessing algorithm is optimized,and the advantage of parallel arithmetic of FPGA is given full play to;the lane-line recognition algorithm based on gradient features is proposed,which can ensure a high recognition rate for lane lines with different widths,lengths and curvatures;furthermore,the gradient operator feedback control mechanism is introduced to adjust thresholds in real time according to different external environments,which can adapt to different driving scenes and output stable and correct recognition results.(3)The detection and classification of traffic signs.In this paper,a real-time recognition algorithm for traffic signs based on machine learning is proposed.The Haar feature-based Adaboost cascade classifier detection algorithm and random grouping convolutional neural network algorithm are adopted for detection and classification of traffic signs respectively,and the modified Haar feature and CART-based weak classifier structure and random grouping convolutional neural network are proposed in combination with the characteristics of FPGA,which improves the real-time performance of the algorithms on the premise of ensuring the recognition accuracy.(4)The lane information modeling technology based on incremental digital maps.In this paper,a method of lane information modeling based on incremental digital maps is proposed,the establishment of incremental digital maps,map matching of positioning information and transformation methods of coordinate systems in maps are introduced,the road information modelling and incrementally updating of digital maps are performed with the multi-information fusion algorithm,and a multi-information fusion software and hardware platform for incremental maps is built,which achieves the real-time update of incremental digital maps and improves the timeliness and accuracy of digital maps.