Design Of Embedded Panoramic Vision System Based On DM642

Panoramic image that covers more than a hemispherical vision field is a pop research issue of environment perception.Because of especial advantage of vision field,it has wide prospects of application on many fields such as military and civilian affairs.Generally,PC is used as a platform for panoramic image processing,which features high cost,big size,and high energy consumption and does not fit the need of high reliability.With the development of embedded system technology,DSP chips are provided with more and more processing capabilities,and it is advantaged to process vision images by an embedded system. On the ground of analyzing characters of panoramic vision system and using the successful application of DSP chips in multimedia for reference,this paper represents the idea of combining panoramic vision and embedded system and introduces the design and implementation of Embedded Panoramic Vision System with a TMS320DM642 chip as the core.Firstly,a modular hardware is designed aiming at the characters of panoramic vision and the requirements of application.The hardware consists of DM642 core system,PAL video extended board and high resolution video extended board.The system can not only process PAL video panoramic image but also high resolution video panoramic image from a CameraLink interfaced camera, and this increases the value of application.After the hardware platform,the software of system is designed including DSP/BIOS embedded real-time operating system,video port driver and software framework based on TI's RF5 which builds a good software platform for algorithms.The main function of Embedded Panoramic Vision System is to revert a panoramic image to a unwrapped image for human's watching habit,it is convenient for monitoring and further vision algorithms.In this paper,the cylinder unwrapping,fast unwrapping,and perspective unwrapping algorithms are studied.These algorithms are improved aiming at the characters of DSP in methods of seeking table and block prefetching.Proved by the test data,the improved unwrapping algorithmsThis system features small size,high stability,and real-time processing.It can be used in monitoring and builds a good platform for applications on target recognition and tracking,and autonomous mobile robot vision positioning and navigation.