Feasible Area Detection Based On Stereo Vision

With the rapid development of computer vision and artificial intelligence technology,advanced driver assistance systems and even driverless technology have become the research hotspot in the field of intelligent travel.In order to ensure the safe driving of intelligent vehicles,it is necessary to detect the feasible area of the road ahead of the vehicle,and use the three-dimensional information of the road to assist the vehicle in path planning and obstacle avoidance.In this paper,a binocular and multi-purpose stereo vision system is built to detect the feasible area of the campus road surface.The feasible area detection system proposed in this paper has the characteristics of low cost and low power consumption,which can well adapt to the complex outdoor environment and lay a good foundation for traffic planning and obstacle avoidance.The system installed on the campus patrol vehicle can help the security personnel complete the campus patrol tasks.First of all,this system is to use two-dimensional image information to obtain the feasible area of the target road.The basic model of the camera in stereo vision,including: small hole imaging model,binocular model and nonlinear distortion model.In this paper,the commonly used Zhang Youzheng calibration method is used to calibrate the camera.The Zhang Youzheng calibration method does not require special calibration objects.It only needs to use the different relative position information between the camera and the checkerboard to solve the internal and external parameters of the camera.In this paper,the Bouguet correction method is used to correct the image acquired by the camera,so that the two camera planes are parallel to each other to reduce the two-dimensional image matching to onedimensional matching,which reduces the computational complexity.In this paper,an improved weight-based semi-global stereo matching(SGBM)algorithm is used for stereo matching to obtain the disparity map of the target scene,and then the U-V disparity map method is used to detect the feasible area of the road.Secondly,based on the traditional semi-global block matching algorithm,this paper introduces the concept of weights,assigns a window to each pixel when performing cost calculation,and assigns weight to each cost in the window for aggregation.The weight is determined by the space and color distance between the point to be matched and its neighboring pixels.The addition of a window with weights effectively reduces the computational complexity and guarantees the quality of the disparity map.When performing U-V disparity maps for road detection,constraints are added for V view and line detection,which improves the detection quality.Finally,this paper builds an acquisition processing platform based on binocular vision,and tests the actual campus road surface,verifies the practicability and real-time performance of the algorithm,and statistics the recognition rate of the algorithm.After binocular vision is completed,the trinocular stereo vision is explored and studied on this basis.The advantage of trinocular vision is that it can effectively reduce the occlusion and mismatch problems in binocular vision.The quality of the disparity map has been further improved.