| Underground mine rescue robot plays a big role in the field of mine rescue. However, underground mine rescue is bad at self-navigation, which limits its application in underground mine rescue. Stereo vision technology can greatly enhance the intelligent level of rescue robot and disengage from any dependence on human. Based on the principle of real time and reliability for underground mine rescue, this work centers on path planning and stereo matching.For the study of path planning, two path planning algorithms were put forward to match the narrow tunnel workspace. One algorithm is designed based on the artificial potential field. In this algorithm, the potential field model which can accurately reflect the outline of the forbidden region was derived by Cumulative Distribution Function(CDF) firstly, and the concrete method and principle of establishing the potential field are researched by making use of Logistic Distribution. Then in order to propose a fast path planning method based on the new potential field, we analyzed the essential characteristic of Quasi-Geodesic and problems when apply to the new potential field, an improved Quasi-Geodesic method is proposed. At last to overcome the inherent shortcomings for APF, a two-way heuristic method is presented. Simulation results show that this algorithm has good performances for computational speed and robust. Another algorithm is inspired from Level Set. Firstly, a method based on the Nearest Boundary Point Passing is proposed for computing the distance level set, and for the purpose of improving the speed of the algorithm, memory pool and binary sort tree are used for the implementation of algorithm. Then for the sake of getting a smoother path, we analyzed the oscillation of path obtained by traditional level set method, so the speed expression of Eikonal function is computed by making use of the distance level set and the new cost between grid point and destination point can be obtained. According to the cost, the initial path can be computed by a synthesis method which integrates the rk4 and bilinear interpolation. Last for optimizing the path, decreasing the length of path and easy to control the distance between the path and forbidden region, a path optimum algorithm called elastic particles method is proposed according to active contour. Simulation results indicate that this method is fast and reliable.Two stereo matching algorithms based on local matching method are proposed. One is derived from adaptive support-weight approach with fixed size window and LBP with image pyramiding. This method has a better matching effect. Another is designed from variable window approach and DP. This method has a shorter execution time. In the first approach, line based matching window is adopted to decrease execution time firstly; Then, in order to make good use of the limited pixel information and enhance the matching effect, a new adaptive-weight mode is designed according to the segmentation matching theory. In this model, neural network is used for computing spatial weight, and Mean-Shift algorithm is put forward for calculating intensity weight, measure function proposed by Birchfield and Census transform are used for cost aggregation. Next, Loopy Belief Propagation is the optimization method in this dissertation. Finally, an iterative left-right consistency checking is proposed for disparity refinement. Simulation results indicated that compared with Yoon’s method and improved Yoon’s method, error matching rate of our algorithm is lowest. The influence of parameters on the matching effect is identified by simulation and reference values are given. The initial energy of LBP algorithm is reduced effectively, which means it can obtain a better disparity map with less iteration. The results of this algorithm is 5(7.1.2014) in Middlebury Stereo Evaluation, this showed up good matching effect of this algorithm. In the second approach, Firstly, in order to enhance the matching effect, edge-stopping anisotropic diffusion is adopted to preprocess the image. Then, seeking strategy of the final matching window is improved, a step of searching matching window for the right image is removed, so the execution time is reduced significantly. Next, similar to the first stereo matching algorithm, measure function proposed by Birchfield and Census transform are also used for cost aggregation. At last, dynamic programming is employed for calculating the disparity map. For the sake of reducing execution time, min convolution is introduced to improve the dynamic programming algorithm. Simulation results illustrated improvements for filtering method and cost aggregation method can effectively reduce error matching rate and execution time is greatly reduced by adopting new searching method and improved dynamic programming method. Simulation result of this algorithm is 37(2.18.2014) in Middlebury Stereo Evaluation, and this matching method has high accuracy while the execution time is low. |
PDF Full Text Request