The Research On Algorithm Of Mobile Robot Vision Location And Simultaneous Mapping

Mobile robot vison is one of the most important technologies of realizing robotinelligentize, and it is also a key reseach that paid close attention. Robot vision mainlythrough visual system to obtain two dimension image,percept and recover three dimensionobject’s posture, relative position and geometry information and so on, then describe andexplain around environment, and transmit the obtained information to robot, and the robotsystem would make the determine. This paper research visual matching and SLAMalgorithm, and do some improvement to the original.The main content as follows:Firstly, this paper summarizes the robot vision’s background and significance, thenintroduces the robot inelligentize technology, the development in demestic and overseasand visual system’s current situation.Secondly, it introduces the conception of robot location and some common usemethods, and then states the relation of world coordination, vidicon coordination andimage coordination. According to the image information, we will explain the method offeature points’ detection, and analysis the idea of SLAM.Thirdly, for the disadvantage of using image pixel difference of derivation methodand ignoring geometric information descriptors of the traditional SIFT algorithm, thispaper uses Gabor derivative filter for its improvement. Taking the advantage of thecharacteristic of frequency and direction of the Gabor filter, the local frequency anddirection are selected for filtering function. Through conducting convolution of the Gaborfilter function and the matching image, and acquiring image feature informationdescription, it removes the noise effectively. Combined with the belief propagationalgorithm, the belief of feature points fuse into the message transfer process via the localmapping between feature points. It uses the space constraint relation between featurepoints mapping in the image, and then iteratively calculates critical point in thecorresponding image. Experiment result shows the improved algorithm improve theaccuracy of nearly10%in the mobile robot of image registration.Lastly, it puts forward the visual SLAM method based on the improved SIFT featuretracking.and extracts and tracks the environmental feature points collection,then applies the extended Kalman filter algorithm to estimating the next time robot posture accordingto observation model.Combined with visual SLAM method,the robot obtains theenvironmental information,extracts its feature, and locates the route marking’s location,then updates robot posture and robot’s map. The experiment result declare it is a effectiveand current method.