Robot Obstacle Recognition And Target Tracking Based On Vision

With the rapid development of robot technology, the application field of mobile robot is more and more widely. Meanwhile, robot obstacle recognition and target tracking algorithm based on vision is the attention of people, and it has become one of the highlights in scientific research of the mobile robot in recent years. Compared with other non visual sensor research, based on visual method has some advantages, such as good concealment, rich information and target information complete, etc.At present, many scholars at home and abroad did the researches on the obstacle recognition and target tracking, were mainly based on the binocular vision. There existed some problems as computational complexity, small visual range and vulnerable to polysemy. The experiment process of the obstacles recognition and target tracking, were mostly the algorithm of simulation results, and the accuracy and real-time were relatively poor. In the process of robot vision research, this paper, in consideration of the superior features of the monocular vision, designed the corresponding algorithm models and the experimental schemes, and did the experiments on the real robot. And it had resolved the problems of the obstacle recognition and target tracking accuracy, real-time, computational complexity, etc. The full paper were divided into three modules of the obstacle recognition, visual range and target tracking with monocular vision robot, and had done the in-depth research on them,In obstacle recognition, chose the appropriate image processing algorithm for gray processing, denoising in smooth way and threshold segmentation. Then propose a method based on HSI color space and combined it with the SIFT algorithm, namely first use the HSI color space to identify the obstacles in rough, then use the improved SIFT algorithm to accurately identify the obstacles;In monocular visual range, analyzed camera model and given some methods for related parameters calculation. According to the geometric distance measurement, deduced the distance measurement model of the experimental process. Through the experiment in this paper, the ranging algorithm can meet the needs of accuracy and real-time performance. At last, discussed and analyzed the error precision of the monocular vision range;In target tracking, mainly aiming at moving target object, designed a target tracking scheme. Position the target objects by using the moving target detection and object recognition methods. Then, the Kalman filtering algorithm was applied to this target tracking system, which could forecast the identified target objects’ motion state. According to the robot’s motion control system, made the corresponding control decision, and successfully realized the real-time tracking of moving targets.Through MT-R robot programming experiments, this paper had completed the task of the obstacle recognition, monocular visual range and target tracking based on monocular vision. Monocular vision of the experimental results showed that the research on the premise of meet the real-time could accurately identify the obstacle, and track moving target object accurately. At the same time, this article also had carried on the analysis and evaluation of monocular vision experimental results, and the corresponding improvement schemes were added to this paper.