Research On Key Technology Of Unmanned Vehicle Perception System

Unmanned technology is an important development project in today’s cutting-edge science and technology, it is for many aspects of social and economic development, national defense and science and technology development are of great influence.Unmanned technology involves cognitive science, sensor technology, computer technology, artificial intelligence and vehicle engineering interdisciplinary content, includes research of basic theory and breakthroughs of key technology, but also involves a lot of engineering design and implementation issues.In the constituent unit of the unmanned technology, environmental perception system is a vital part of entire unmanned system and guarantee the security and intelligence of the unmanned vehicle. The detection of road information is the core issue of environmental perception systems, road information based on the lane detection technology, obstacle detection technology and road boundary detection technology is the key technology of unmanned vehicle sensing system. It is a necessary precondition of autonomous unmanned vehicle safety driving, but also been the focus and difficulty of research. The research aiming at the key technology of unmanned vehicle environmental perception system, including sensors system calibration, lane detection, obstacle detection and road boundary detection of several aspects, and study in actual engineering design and implementation of some specific technical means. The specific research contents and innovations of this thesis include the following aspects:1) The domestic and foreign development situation of unmanned vehicle is investigated, and the realization method of the environmental sensing system technology is analyzed and compared, and the key problems of the system are put forward. On this basis, the design idea of intelligent pioneer "series of unmanned vehicle system of the environmental perception and the research content of this thesis are put forward.2) The calibration method of laser sensor and vision sensor is studied, and the self calibration and combined calibration method of laser sensor and vision sensor is designed and realized. Through the design of visual interface to obtain the position of the feature points, reduce the laser sensor and vision sensor data feature obtaining and matching the difficulty, improve the calibration efficiency and accuracy, provide the conditions for the treatment of the following algorithm.3) Given the problems of less single-image information and poor anti-interference ability, a lane detection algorithm based on temporal-spatial information matching and fusion by reverse perspective transformation is proposed in this study. Algorithm combined with INS information, using the camera inverse perspective transform algorithm and SAC-IA algorithm, to achieve the characteristics of the lane line space matching fusion. Compared the traditional lane line detection algorithm, obtained greater scope and amount of data is more abundant in the lane line feature data, and using linear clustering algorithm based on density to realize the lane line feature data clustering and filtering the interference; Introducing prediction-tracking model to further improve the accuracy of detection. Through experimental results and statistical analysis show that the algorithm enhances the lane line detection of anti-jamming capability, and the algorithm has high accuracy and robustness.4) Requirements for unmanned vehicle systems obstacle detection reliability and high stability problem is proposed based on three-dimensional laser radar obstacle detection algorithm for four-dimensional spatial filtering. Algorithm using spatial domain analysis method of 3D laser radar data is processed to detect obstacles, to avoid the traditional raster map maximum minimum height difference obstacle detection method based on due to the segmentation caused by undetected problem. Algorithm also changed the traditional obstacle detection method only in 3D space for processing and transformation to the time domain of the four-dimensional space to perform filtering processing, can effectively reduce the unmanned vehicle bumps, etc. caused by "false alarm" problem and improve the stability of detection. Experimental results and statistics show that the algorithm of the road environment in a variety of different types of common obstacle detection performance has good performance, high reliability and stability.5) According to the extraction of road surface in complex road environment, a new method of road boundary detection algorithm based on B spline model is proposed. In the algorithm, the feature points of road boundary are extracted by using the method of 3D laser neighborhood curvature gradient, which can be used to extract the feature points of the obvious feature point road boundary. Adaptive circle search algorithm is used to obtain the candidate road boundary data, and the FCM algorithm is used to realize the road boundary feature points clustering to filter the interference. The algorithm also introduces the knowledge of the road model, and the B spline model is used to fit the road boundary, which improves the robustness of the external influences such as disturbance and discontinuity. The algorithm overcomes the previous algorithm of road boundary shape and height requirements are relatively high, and can adapt to a variety of road environment. Experimental and statistical analysis results also show that the algorithm is stable and reliable, and can meet the needs of environment modeling and path planning of unmanned vehicle in complex road scenes.