Research On Visual Detection And Tracking Of Mobile Robots

The vision system is one of the important parts in the mobile robot system,it provides information about the position and the motion parameters of mobile robots for detecting environment around. In the background of general mobile robot system experimental platform, this paper devoted to object detection and tracking problem, based on global camera and local robot camera respectively. The main content of the paper was listed as follows:First, for the specific characters of vision detection and tracking problem, and experimental environment, the pre-processing methods including image-gathering, noise removing and image segmentation are studied. And then, usual algorithm for object detection is compared and analyzed. Considering the shortage of existing methods, an improved motion history image object detection algorithm is designed for less complexity and high efficiency. Kalman Filter is combined to improve detection veracity and efficiency. Mobile robot localization and tracking algorithm based on global vision is also proposed, achieving mobile robot precise detection and tracking by global camera.Next, mobile robot location and tracking algorithm based on local vision system is proposed. Designed combined color signs for mobile robot system by the precision requirements of multiple robots recognition and detection under many conditions. And according to the characters of color signs, a least squares-based ellipse fitting method and D-P based rectangle detecting algorithm are synthesized to achieve robot recognition and subpixel level feature points detection. Afterwards, a multi-resolution Lucas-Kanade optical flow method is used to multiple objects tracking on the basis of corner features, which results show stable detection and tracking for mobile robots in moving scenes.At last, on the basis of the forementioned study, a vision system for mobile robots detection and tracking is established, which provides an experimental platform to validate relative algorithms. The design and implementation are researched in detail, including system main frame,main function of each model,estimation methods and their limitations. Meanwhile, according to the experimental result under different conditions, system performance analysis is given about system veracity, real time capabilities, algorithm robustness, wide view field, tracking precision and expansibility and so on. The experimental results indicate that system performance satisfies system design requirements, and system framework and algorithms exhibit good robustness.