Research And Realization Of Artificial Landmark Recognition Algorithm Based On DSP

AGV (Automated Guided Vehicle) has been widely used in our daily life and all industry sectors since its advent. At present, as one of the guiding methods used in AGV, machine vision navigation has not been expanded broadly because of its technical difficulty. But rich information, great flexibility and immune to electromagnetic interference make vision navigation become the main development direction of future guiding technologies. More importantly, using artificial landmark as the guiding identification is getting more and more attention.Firstly, in this paper, some artificial landmarks designed by domestic or foreign scholars are investigated. Then, with the obtained information, a novel artificial landmark pattern is proposed which consist of guiding light, black band and character. Guiding light and black band is used to locate the landmark. The guiding light plays an important role in fast locating to artificial landmark. At last, both principal component analysis method and minimum distance classifier are used to recognize the character in the artificial landmark.After the overall design was completed, the algorithms are simulated on Matlab and realized on DSP. On Matlab, the recognition algorithm for guiding light is simulated and the mean vector of training samples, feature vectors of training sample, the optimal PC numbers and PCA projection matrix that corresponding to the highest recognition rate are also generated on Matlab. Then, the recognition and localization to guiding light, localization to characters based on invariance of cross-ratio, segmentation to characters based on projection method, dimension reduction to the image using the mean vector of training samples and projection matrix and the recognition to character using the minimum distance classifier are completed on DSP.ICETEK-DM6437-B evaluation module is the hardware development platform, whose core CPU is TMS320DM6437.The experiment showed that the recognition rate of the algorithm I designed is above90%in the limited angle and distance range, and has high recognition rate and practicality.