Gait Recognition Using Deep Convolutional Features And Hash Coding


Pa, Pa Min (2020) Gait Recognition Using Deep Convolutional Features And Hash Coding. Masters thesis, Multimedia University.

Full text not available from this repository.
Official URL:


Human gait recognition is the tracking of the people by the way theywalk and the recognizing of human motion. It has gained attention because of its non-invasive and unobtrusive behaviors as well as its applicability in the various areas. Gait recognition is perceived as the most promising biometric approach for the next coming decades, especially because of their efficiency and applicability in a surveillance system. Gait recognition becomes a challenging task in improving accuracy because of its covariant factors such as views, speed, and walking conditions. Due to the recent increase in gait data across surveillance systems, a rapid search for the required data becomes an emerging need. Researchers tend to pay less attention in addressing the issue of retrieval problem. The gait retrieval problem is to search the target individual in large datasets in an accurate manner. Hence, the objective of this study is to propose an efficient gait recognition framework to address the gait retrieval problem in a largescale dataset. The deep gait retrieval hashing model (DGRH) is proposed to retrieve the similar gait person from the large-scale dataset given by the query subject. The DGRH model uses the deep supervised hashing approach, with a convolutional neural network. Hence, the suitable CNN architecture for the gait feature extraction is proposed. The three CNN architectures with different activation functions are studied and their performances are analyzed in gait classification and verification framework. The classification model, with different architectures is evaluated in terms of correct classification rate (CCR). For the verification framework, the proposed CNN architectures are built in the Siamese network to address the one-to-one association. The performances are evaluated against the equal error rate (EER). The DGRH model uses the convolutional neural network to learn the features representation and the last layer of the architecture generated the hash codes. The hash function is learned by optimizing the classification loss, which is the responsibility for the similaritypreserving with gait labels and quantization loss for the hash quality. The ability of CNN is able to generate quality hash codes. The gait retrieval was performed withthe Hamming distance between the database hash codes and query hash codes. The proposed framework is evaluated against the three public datasets (CASIA-B, OUISIR-Large Population Dataset, OUISIR Multi-view Large Population Dataset) with cross-walking conditions in terms of mean average precision (MAP) and precision curves based on the top-returned images.

Item Type: Thesis (Masters)
Additional Information: Call No: TK7882.P3 P37 2020
Uncontrolled Keywords: Pattern recognition systems
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800-8360 Electronics
Divisions: Faculty of Information Science and Technology (FIST)
Depositing User: Ms Nurul Iqtiani Ahmad
Date Deposited: 22 May 2023 06:06
Last Modified: 22 May 2023 06:06


Downloads per month over past year

View ItemEdit (login required)