Study On Gait Recognition Method Based On Lower Limb Centripetal Acceleration Under Monocular Vision

In modern society, with the development of science and technology and the increasingof information, the security problem of public sensitive place holds researchers’attention invarious fields, as a result, various biometrics are springing up. As a new method, theidentity recognition technology based on measurement of plantar pressure and analysis ofgait has some special advantages, such as difficult to hide, no damage identification andlong-range identification, these factors make this method cause more and more attention inthese years. However, the tactile recognition system based on the plantar pressuremeasurement can’t be widely used because of high costs of equipment, but the equipment ofvisual identification analysis based on the analysis of lower limb centripetal acceleration isrelatively simple and can be used as a supplementary means, and also has practicalapplication value.In order to improve the performance of existing identification system, a novelidentification method based on the combination of circular movement and centripetalacceleration from vision aspect was proposed, the feasibility of the recognition principlewas clarified and the lower limb mechanical model was established, followed experimentalverification in single view. The major work of the dissertation is as follows:Firstly, the merits and drawbacks of these existing identification means at home andabroad were analyzed, and also the research status of recognition technology. A gaitrecognition method based on human lower limb centripetal acceleration was proposed inthis dissertation, and a general mechanical model of human lower limb movement wasestablished. The feasibility of the identification method based on motion model was alsoanalyzed in this dissertation.Secondly, the walking gait image sequence processing work was finished in singleview, a novel background modeling technology on account of the combination of framedifference and average in time domain. On this basis, the moving foreground segmentation,image filtering, image de-noising and morphological processing were launched.Then the image sequences feature periodic extraction was finished under a single view,single leg’s key posture motion was analyzed according to the characteristics of periodicityand symmetry during walking. The impact of human upper limb during walking is ignoredand the lower limb movement is regarded as a circular motion, the movement of the thigh isregarded as hinged lever rotation around the hip joint, and the leg movement is regarded asa circular motion around the knee joint, the joint point moving trace is a part of the circle.Using the theory of circular motion, the hip, knee and ankle joint point of moving humanlower limb were extracted, the lower limb mechanics modeling of the movement at a particular posture was established and the centripetal acceleration was calculated.Finally, according to the platform that established before, the lower limb motion imagesequences’ features of different people were extracted and the motion parameters andstatistics information were finished. Then a small experimental database was built,according to gathering some pedestrian movement information again and the comparisonwith the database, the identification experiment was finished. The test achievementindicates the feasibility of the identification method used in this dissertation.