A Fuzzy based mathematical model to motion detection with monocular vision

Abstract

Vision based decision making and obstacle avoidance of a mobile robot are the most attracted areas of the domains of Computer Vision and Robotics respectively. The link between them is, Computer Vision, more specifically the vision for intelligent machines is heavily used for mobile robots as a primary sensory input to perceive the external world. Vision based obstacle avoidance is one of the major areas of study Robotics. Estimating and predicting the motion behavior of a dynamic object with a single camera, known as monocular vision, is a real challenge. We postulated this can be done by analyzing a sequence of image frames extracted from a live video stream. But, these analytical techniques must be extremely fast in real time processing, because the decisions drawn within reasonably short response time are the only factors to ensure the safety of the robot. Therefore, we postulated a fuzzy mathematical model - an artificial intelligence approach to perform that task, which has a significant impact in terms of simplicity (reduced complexity) together with efficiency (minimized computational overhead : resource consumption), rather than conventional complicated mathematical modeling. The evaluation results of the all major modules of the software artifact developed to test the hypothesis of the research, namely Image Processing Module, Fuzzy Mathematical Model exhibit the validity of the concept postulated. For example, average accuracy of the Image Processing Module and the Fuzzy Mathematical Model (Simulator version) are 89% and 92% respectively. Finally, it is proven with strong evidences that, fuzzy mathematical modeling is appropriate to recover motion even from simple inputs such as apparent size variation, while monocular vision is adequate perceive to capture those primary inputs. Therefore the ultimate milestone of the research, developing an artifact (a strategy) to test the conceptually postulated theory in both theoretical (perfect i.e. simulation environments) and practical (real world noisy environments) aspects, has been achieved successfully.