The Research Of Computer Vision In Excavator Robot’s Target Recognition And Location

Acting as an important equipment in engineering construction, excavators play a vital rolein the construction of industrial and civil architecture、transportation、water conservancy andpower engineering construction、mining extraction and military engineering construction,etc. Butin order to overcome its problem of high labor intensity、hard working condition、some areas ofinconvenience and so on, the researches on robotic excavator have become a hotspot both athome and abroad. In the past thirty years, with the further development of the computer visiontechnology and microelectronic technology, building a fully functional vision system forexcavator robot has become another hotspot among the scholars at present. In this paper, afteranalysing the history and status of the computer vision applied in excavator robot’s targetrecognition and location, the author takes the physic model of Buliding industry Share’s XG806hydraulic excavator as the reconstructive object and builds virsual system with target recognitionand location function for it on the basis of having remoulded its hudraulic system intoelectro-hydraulic proportional system, the results improve the excavator’s intelligence further.The main work is as follows:1.The experiment platform of excavator robot is proposed. The major parts of its visionsystem are selected and located, and then the identification and location scheme of target aredetermined.2.The scale invariant feature transform(SIFT) image matching algorithm is proposed basedon analyzing the current target matching algorithms and location technologies which have beenapplied to the excavator robot’s visual system. Experiments verify the superiority of SIFTcomparing with the traditional algorithms and the higher versatility and robustness can beobtained when apply it to the excavatior robot’s target recognition.3.The original SIFT feature descriptor in another equivalent data representation are mappedinto a high-dimensional feature space by gaussian kernel function so as to make the originaldatas linearly separable, and then in this space the dates which can represent the characteristicsof the original datas are determined by calculating the contribution rate which dues to the featuredescriptor contributes to the total feature descriptor. As a result, the datas that exist in the "noise"and "redundant" are cancelled. Experiments show that the improved SIFT image matchingalgorithm shortens the time of image matching and obtains higher matching accuracy. 4.To solve the problems that the visual control system of the excavator robot which is builtup by Labview cann’t identify and locate the targets all the time when the targets take place therotation、scaling or the light condition has changed, The improved SIFT M files are invoked intoLabview through the Matlab script and Read from spreadsheet file VI nodes, which realizes theseamless connection of Labview and Matlab. Experiments show that: the Labview and Matlabhybrid programming for the excavator robot’s visual control system solves the problems exitedin the old Labview visual system and is more perfect and powerful.5.For the convenience of verification test, the robot manipulator which has the samefuncitons as excavator robot’s working device was selected to debug and validate. Experimentsshow that: building a virsual system with target recognition and location function for excavatorrobot can improve its intelligence.