Footprint Feature Analysis And Application Research Based On Flexible Sensor Array

With the application of the footprint recognition, it is getting more and more attention that the footprint feature analysis is based on the flexible array sensor. The characteristics of plantar pressure data is robust for cover, size and changes in the background. The plantar pressure data contain rich information about the object’s gait posture, gait habit and footprint. So it is widely used in some fields, such as clinical medicine, motion analysis, criminal investigation and security, identity recognition and so on. It can provide quantitative scientific basis for the decision in these field, where footprint feature is extracted effectively.In this paper, based on the domestic and foreign research about footprint feature extracting, analysising and recognizing, the footprint recognition is researched with the techniques including feature extraction, feature selection and multi-source feature fusion. And it is also applied to footprint tracking. We construct the static database of plantar pressure under standing and the dynamic database of plantar pressure under walking. The static database includes two sets of data under certain conditions, in order to study the influence of some factors. The dynamic database of plantar pressure consists of two parts. The one is plantar pressure data of different objects by walking with four categories, of which there are barefoot walking with often speed, barefoot with quickly speed, shoes walking with often speed, and shoes walking with quickly speed. And the other data set is plantar pressure data, which is collected from many people walking at the same time. The main contents and innovations are as follows:1. Extracting of biomechanics feature, morphology feature and texture feature for plantar pressure data. The regional dynamics features were extracted in order to obtain the relative distribution of plantar pressure. Meanwhile the morphology feature and the texture feature are extracted. The morphology feature includes shape features and Hu- moment features, and the texture features is quadratic statistical coefficients by gray-gradient co-occurrence matrix. The stability of features is investigated by calculating the intraclass correlation coefficient (ICC) and coefficient of variation (C.V). The experimental results show that the most features have good stability and uniqueness.2. Proposing the feature selection method of multi-footprint (MFFS). Firstly, the biomechanics feature subset is selected by the Criterion of Correlation and the first search algorithm. Secondly, the coefficient of feature weight is calculated according to the importance of feature. Through the MFFS method for classification, the largest contribution features are retained. The feature selection of various forms plantar pressure data is also researched, which includes 3 types of incomplete data from forefoot, heel, and their separation. The results indicat that, the accuracy of identity recognition of MFFS is higher, and the MFFS method can greatly reduce the feature dimension of the footprint data.3. Proposing the fusion method of footprint features based on the confidence of feature. The fusion method of the evidence synthesis based on confidence of feature is used. The evidence set is constructed according to the new evidence combination rule, which enhances the feature of high confidence, and weakens the feature of low confidence. For the footprint tracking system, the results of D-S evidence synthesis rule are compared with those of the new evidence synthesis rule, which shows that it is better by the new evidence synthesis rule.4. Realizing the the footprint tracking experiment based on the flexible array sensor in order to realize the multi people tracks. And the amount of data is reduced by prismatic effective area. The result is obtained according to the MFFS and the new fusion rule. Researching footprint tracking of two objects and many objects, this has cross point or no cross point. The result shows that, the accuracy of recognition is good, meanwhile the accuracy of recognition is improved.