Research On Image Processing Techniques For Intelligent Excavator’s Vision System

With the lack of terrestrial resources as well as the continuous advancement of technology, human beings exploit into a wider area such as the sea and space. Due to restrictions of the environmental conditions, underwater operations and space development are very difficult to achieve manually. Furthermore, with the complexity of handling radioactive substances in the poor and dangerous environment, the operator intervention is inconvenience or the safety for the operator can not be guaranteed, so the intelligent excavator is a substitute to complete these complex and dangerous works. Therefore, the research on intelligent excavator has important theoretical value and engineering significance. The visual system as a necessary to achieve intelligent excavator, its importance is self-evident.This paper makes a systematic and in-depth theoretical analysis for several image processing technical of the visual system to the application of an intelligent excavator, some improved algorithms have been proposed, at the same time, the application of the visual system has been expounded aimed at the concrete excavator’s work. The main works in this paper are as follows:(1) Analyzing the traditional Chan Vese image segmentation model which can not smooth the noise and preserve the weak edge at the same time, an improved segmentation algorithm is proposed. It introduces the anisotropic smoothing term to the segmentation model, and uses of edge detection operator to mount the initial contour. The experiments results show that, the noise is filtered and the weak edges are preserved at the same time. The capability to capture the outline on the target is enhanced. Meanwhile the initial contour mount method can speed up the operation speed.(2) Aiming at the current global contour matching method which can not match under the occluded case, a local contour matching algorithm is proposed which is based on section contour curvature. First, compute the curvature of each point on the contour, and choose the candidate points which meet the threshold condition. After that segment the contour based on the points. The feature vectors are constructed by using these sub-contours. The variance is used for measuring the similarity between the model contour and the target contour. The differential evolution algorithm is adopted to deal with the issue of optimization. The experiments results show that, the improved method can cope with that the global contour metching method can not obtain the metching result when the contour is occluded, and using the differential evolution algorithm can speed up the operation, meanwhile the optimal result can be obtained.(3) After analyzing the problem of the large errors in optical flow estimation which is caused by low detection precision on occluded regions, the improved occlusion detection algorithm is put forward, and use for optical flow estimation. Use the joint features in the image, to detect the occluded regions, and to cope with the motion discontinuity by bilateral filtering. The results show that, the improved optical flow estimation algorithm can enhance the accuracy and robustness of the estimation on occluded regions, and can acquire the more precision optical flow information.(4) Aiming at the the problem of easily locate the error position by the classic Mean Shift which is caused by no the pixels’ space information in the feature space representation when there are similar colors in the background around the tracking object, an improved object tracking algorithm is proposed. At first, the target model region is segmented into overlapped square, and their spatial histograms are computed which has introduced pixels’spatial information into. So the weakness of the weighted color histograms is overcomed and the accuracy is enhanced. The experiment results show that the improved algorithm can track the object more robust, accurately and quickly.(5) At the last, aiming at the turn and unload procedure of the excavator, the application of the improved slgorithms is discussed in the actual work. Meanwhile, the feasibility is testeland verifiaiby the experiment. And the issue for fusing the visual signals in intelligent control system on the intelligent excavator is discussed.