The Gait Recognition Based On Convolutional Neural Network And Plantar Pressure Information

As a biometric feature, gait plays a pivotal role in identity recognition. Compared with fingerprint, face, iris, etc, gait has many characteristics, such as inimitability, long-distance observation, and independence on the tester when collecting gait data. So gait recognition becomes a reseach hot-spot in recent years. Now there are two ways of gait recognition:the earlier developed gait recognition technology based on video/image, develop earlier, has made certain progress. But the recognition effect is vulnerable to the interference of external environment, such as complex background and weather. The gait recognition based on foot pressure information, do not require the intentional cooperation of testers when collecting data, so the acquisition equipment is easy to hide and has very good anti-counterfeiting ability. Although some existing algorithm achieves good recognition results, but the large amount of calculation for feature extraction, feature parameters of complex adjust is needed. To address these issues, this thesis studies a new gait recognition method based on convolutional neural network and plantar pressure information. The main work and research results are summarized as follows:1. In terms of establishment of the gait database, this thesis analyzes the basic condition and data collection method at pipeline, then designs the data collection process, According to the obtained process, we collect standing and three walking data with different speed. We perform preprocessing on the obtained data in this thesis.2. In terms of feature extraction, on the basis of analyzing the traditional feature extraction method, this thesis proposes a convolutional neural network model that can reduce the calculation of traditional feature extraction algorithm. By integrating local features of small area we can obtain the convolution feature of the original image. Single convolutional neural network model and double convolutional neural network model are used to get S-CNN and D-CNN feature. We compare the these two kind of features and find that the network depth has influence on the recognition accuracy.3. This thesis also introduces a kind of dividing method. We extract the interested region, and give the weights of each region according to each region on the whole image role, and finally obtain integrated image features. As the input of the convolutional neural network, the integrated features are used to identify the target detection and classification. This thesis applies the model to gait recognition. Experiments demonstrate the effectiveness of the algorithm, and a higher recognition accuracy in static data and dynamic data over other methods.