Study On Adaptive Indoor Visual Positioning Method Based On SUSAN-SURF

With the continuous development of the era of science and technology, demandfor indoor location by people is increased continuously. Visual positioning as a wayof indoor positioning is not affected by complex indoor environment, only with avision sensor to obtain image information, by means of image registration, estimatethe displacement information, so as to realize positioning. The accuracy of thistechnique can reach centimeter level, which gradually attracted more attention frompeople. In this paper, we study a kind of indoor visual locating method based onimage registration.In order to achieve the effective indoor visual positioning method, the mainwork has the following several aspects:1. Do the three mainstream image registration algorithm performance contrastexperiments by SUSAN、SIFT and SURF, the high accuracy of SUSAN algorithmand the quickness of SURF algorithm are verified.2. According to the analysis of the demand of visual location, the accuracy ofSUSAN algorithm is improved: Using the pixel projection method to detect featurepoints, reduces the miss rate, and improve the accuracy of the algorithm in the aspectof extracting. The improved algorithm also compared with the original algorithm toverify the effectiveness through the experiment.3. SURF algorithm on real-time performance is improved: Aiming at thetime-consuming problem with original SURF in the process of matching, this paperuses the KNN to improve the matching efficiency of algorithm and reduces the timecomplexity. Also compared with the original algorithm through the experiment andverified the feasibility of the improved SURF algorithm.4. This paper proposed an effective combination of SUSAN-SURF algorithm,which retains the high efficiency of SURF and the outline information of SUSAN.Write and debug successfully, the algorithm is verified the good real-time, accuracyand adaptability by experiment.5. In order to improve the accuracy of the SUSAN-SURF algorithm further,RANSAC matrix was added in to optimize the algorithm eliminating wrong matching points. Experiments show the good feature point matching results.6. Aiming at problem of poor real-time performance in motion estimation, thispaper use of conventional Kalman filter in recursive calculation to predict theoverlapping area. Only in the overlapping area do the feature point extraction andmatching, increase the adaptability of motion estimation.7. Indoor visual positioning system is developed and experimental verificationis completed. By building software and hardware platform, the experiment resultsshow that the indoor visual positioning can manage this method well.