Positioning Technology, Robot Vision Based On Image Sequences

With the rapid developments of computer software and hardware, image processing technology and robotics technology, computer vision is applied to the field of robotics more and more commonly. In this paper, on the basis of the sequences of images studies on robotic visual localization technology are carried out which detect the moving objects from the video sequences using computer vision and track them to localize their three-dimensional positions, followed by feeding back the position information to the robot, thus guiding it to perform the appropriate action. Under the condition of a static camera the main research on the localization of the moving objects through video images in this paper includes the following parts:Visual localization measurement model. A monocular vision system is introduced to deduce a localization measurement model by means of a linear camera model. Some assumed restrictive conditions are brought forward to simplify the implementation complexity of the system. The mass center is selected to represent the object. According to the mathematical description the object motion model is established, and then motion parameters are derived through the motion characteristics, thus predicting the position and orientation of the object which is converted into the world three-dimensional coordinates as a feedback to the robot by way of two-dimensional image processing.Video image preprocessing. Such as noise elimination, histogram equalization, smooth, morphological operations are processed. Some false targets are removed according to the contour information, thus target movement regions are structured and highlighted to prepare for the follow-up work.Moving object detection and localization. A modified self-adaptive mixture Gaussian model is adopted as the background updating method so as to protect the background model from being influenced at the same time to response to the rapid changes of the background. A color object is tracked with CamShift algorithm in which the movement region obtained through the contour information is set as the initial value of the search window. Afterwards the motion and structure parameters are estimated using the feature-based motion analysis to get the motion path of the moving object. Combined with OpenCV, simulation experiments are designed and experimental results show that the algorithm is demonstrated feasibly to meet the requirement of real-time and accuracy of visual localization.