| Intelligent vehicle has a great application foreground in the field of the military, space exploration and logistics based on the rapid development of computer technology, sensor technology, as well as vision technology. Automatic navigation achieved by means of machine vision has been a hot research area which plays a key role. The environment information gathered around through monocular camera can be used to extract visual information which acquired by image processing algorithm. Simultaneously, obstacles which distribute random can be found out, which obviously contribute to help design a feasible optimal path.In this paper, the vision subsystem consists of monocular camera, which will be applied to study automatic navigation based on machine vision, while obstacle avoidance and path optimization problem should be the primary solution. Thus, the relevant contents include camera calibration, obstacle detection, especially obstacle avoidance and path optimization.The system structure belong to experimental vehicle will be firstly discussed briefly. Reference coordinate system has been introduced to focus on describing imaging model formed from monocular camera, eventually camera calibration will be finished by program made up of OpenCV. Considering decomposing video signal captured by camera into image sequence, which suits further study.This thesis detailedly analyses the obstacle detection algorithm based on image sequence. Many shortcomings can be easily gained once applying frame differentiation method. It is proved that Average Background Model and Gaussian Mixture Model have no advantages as well. Thus, A multi-feature fusion based approach will be put forward to overcome it. Texture feature and edge information were merged with linear weighted method, comparing overall differences between two adjacent frames once updating the background information established by the Gaussian model, obstacle can be distinguished as expected. False obstacles have a great influence to the detection results, A new segmentation approach based on maximum entropy can work here. What’s more that Hough transforms and morphological processing will help identify real obstacles, once connected domain calibration has been done, obstacles can be found out. It suggests that new method has advantage from different comparative experiments.What the scholars care about most lies in optimal path and obstacle avoidance in most of the robot path planning applications. Expanding the detected obstacles properly, and utilizing the distribution of obstacles to establish environment model. Once artificial potential field method has been applied to resolve this problem, certain local extremums will appear easily, which inevitably causes the target inaccessible. Assuming edge detection method can be introduced to make robot walk around the arc, escape from local extremums point, as well as avoid large obstacle. Obviously, it has already increased the path, In this case, A new strategy can be introduced to use. Consider using geometric approach in local extremums point will design a better obstacle avoidance path which helps to escape from local minima areas in short time. Simulation results achieved in many obstacles environment proves that this alternative solution can be more effective by compared with artificial potential field and edge detection method.In addition, global path planning based on ant colony algorithm has been a part of this paper. An accurate workspace established by grid method should come first, when using ant colony algorithm to determine the solution, some problems containing search slowly with more iterative times remain to be solved, which may lead to inefficiency. After making an analysis to many classical algorithms, a new approach can be formed to improve the efficiency. Considering improving heuristic factor, designing new path selection probability, and choosing better access to pheromone updating. It proves that the new method can gain a better path with less time. |
PDF Full Text Request