A Research On Particle Filter Localization Of Mobile Robot With Omnidirectional Vision

In mobile robot applications, it is a fundamental and important requirement that the robot should be able to localize itself accurately within its operating environments. It is also a challenging topic of mobile robot research which has attracted many researchers. Nowadays, vision-based localization of Mobile robots has become a very popular topic of computer vision and mobile robot navigation. This thesis focuses on the Monte Carlo localization of mobile robot with omni-directional vision sensor, then a modified algorithm is proposed to deal with the abundant information of omni-directional vision sensor and to improve the efficiency.Firstly, the function and composing on software of the Monte Carlo localization system of mobile robot with omni-directional vision sensor are introduced, and then the hardware framework of the system is introduced too. The principle of catadioptric omni-directional imaging system and the method to select the size and parameter of hyperboloid mirror are discussed in detail.Extraction algorithm of features in omni-directional image is researched and modified. The Scale Invariant Feature Transform, SIFT, is invariant to image translation, scaling, rotation, and is partially invariant to illumination changes. But, the time of features extracting and matching is huge, and the number of features is much larger then that is needed. So a modified approach based sampling is proposed, compared with the classical SIFT in which features are detected through the whole scale space. The modified SIFT remains the construction of keypoint descriptor, so it can improve efficiency meanwhile keep accurate for feature matching. Important parameter of this modified algorithm is confirmed through experiments, and the algorithm is demonstrated effectual and real-time.The localization of mobile robot is an issue in which the state of system changes with time. Bayesian filter combines the dynamic uncertain estimation of state and the observation, to obtain an accurate state of system with iterative method. It is an appropriate frame for the localization of mobile robot, which is the foundation of probabilistic localization principle. Depending on the Bayesian filter, the classical methods, modified methods and resampling methods of particle filter are studied . Then particle filter is applied to Monte Carlo localization of mobile robot with omni-directional vision sensor. Modified SIFT is used to extract features of omni-directional image, and the mobile robots can localize itself accurately with a lower number of features.