The Target Identification And Self-localization Of NAO Robot On Standard Platform League

In recent years, the research topic of intelligent mobile robots has been increasingly popular with most of the world famous research groups and research scholar. Among this, the problem of target identification and self-localization is the main task to achieve the real intelligent and autonomous robots. Varieties of recognition and localization algorithm are emerging in endlessly, and our aim is the algorithm’s robustness, reliability and real-time. In this paper, these two issues were discussed. After analyzing and comparing various algorithms, we will focus on the Monte Carlo Particle Filter Localization algorithm which is based on the probabilistic framework. By synthesizing the uncertainty factors in the Standard Platform League and the algorithm performance, improve the sample degeneracy phenomenon and particles arid, and then apply the improved algorithm to the NAO robot.First, the core of target identification of the vision-based mobile robots is image processing. The colors used in robot soccer could be accurately classified according to the NAO robot’s visual characteristics. The image processing is split into the two steps region-building and feature extraction with the region-building algorithm and clustering segments algorithm. Then we will give a detailed introduction on extracting the key features, such as detecting lines, detecting the ball and detecting the goal.Second, we discussed the methods of NAO robot self-localization in detail and analyzed the advantages and disadvantages from the traditional Scale Invariant Feature Transform Algorithm to the leading localization methods based on distribution estimation. Considering the actual game, the localization of NAO robot is a non-liner multi-modal probability problem. And the kidnapping will also result in non-Gaussian problem, so we choose the Monte Carlo Particle Filter Localization algorithm, with emphasis on the basic thought and the algorithm flowchart.Thirdly, we analyzed the sample degeneracy phenomenon with the actual execution. Considering the contradictions between the sample number and the computational complexity, importance re-sampling strategy is adopted to solve degeneracy problem. But this method will result in the particles arid, so this paper proposed an improved method using half sampling and half re-sampling strategy.At last, we focus on introducing the execution steps of the improved algorithm based on the Monte Carlo Particle Filter Localization, and applying it to the NAO robot. Finally, the result of experiment is given, and it shows that the improved method could solve the sample degeneracy phenomenon and particles arid and the robot can realize the function of localization.