Mobile Robot Object Recognition With Monocular Vision

Most mobile robots have kinds of intelligent sensors, with which mobile robots acquire environment information, and vision is the most widely used sensor. As the basis of Location, navigation and path planning, object recognition is the hot research domain of mobile robot. In this paper ,we will research the object recognition of mobile robot with monocular vision.Above all, adopting TMS320DM642 and S3C2410 CPU, author designs a embedded vision system to be used for robot object recognition, lucubrating object recognition algorithm based on image matching, in addition, developing an simulation system for mobile object recognition which is built on BP-Neural Network. The specific achievements are as follows.Designing an embedded vision system for object recognition with dual CPU ARM9 and DSP, fulfilling hardware design and introducing fully each circuit module of the vision system. This vision system is working on Windows CE embedded RTOS . In Windows CE, we will give a presentation of software design for mobile's object recognition system.Presenting image segmentation based on threshold and boundary detection, lucubrating the theory and achieving process of histogram bimodality and Otsu algorithm which are based on threshold segmentation algorithm. Introducing image matching algorithm SSDA and NCC based on image gray. Finally, choosing Hausdorff distance as similarity measurement to fulfill image's feature matching.Choosing invariant moment as the feature to identify object, further more, demonstrating its feasibility from theory aspect and experiment aspect. Introducing something about BP-NN, at last, designing a simulation system to run on PC for robot'object recognition with Matlab software.