Person Identification Using Anthropometric and Gait Data from Kinect Sensor
Authors: Virginia Andersson, Ricardo Araujo
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper, we report on experiments using a novel data set composed of 140 individuals walking in front of a Microsoft Kinect sensor. We provide a methodology to extract anthropometric and gait features from this data and show results of applying different machine learning algorithms on subject identification tasks. Focusing on KNN classifiers, we discuss how accuracy varies in different settings, including number of individuals in a gallery, types of attributes used and number of considered neighbors. Finally, we compare the obtained results with other results in the literature, showing that our approach has comparable accuracy for large galleries. |
| Researcher Affiliation | Academia | Virginia O. Andersson and Ricardo M. Araujo PPGC Federal University of Pelotas Rua Gomes Carneiro, 1, Pelotas, RS, Brazil 96010-610 vandersson@inf.ufpel.edu.br, ricardo@inf.ufpel.edu.br |
| Pseudocode | No | The paper describes the methodology and algorithms used in prose, but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper only states that 'The full data set is available at http://ricardoaraujo.net/kinect.', which refers to the data, not the source code for their implementation. There is no explicit statement or link for the source code of their methodology. |
| Open Datasets | Yes | The full data set is available at http://ricardoaraujo.net/kinect. |
| Dataset Splits | Yes | Parameters for each algorithm were set by varying their values while trying to maximize the resulting accuracy using a 10-fold cross-validation (Mitchell 1997) over a smaller random validation subset composed of 20 users. We use 10-fold cross-validation to validate the models with the above parameters i.e. the data set was randomly partitioned in 10 subsets and training was performed ten times, each time leaving one partition out of the training process, which was used for testing; the reported accuracies are the averages of these ten executions. |
| Hardware Specification | No | The paper mentions the 'Kinect device used in our experiments was the 2010 model for the X-Box 360 video-game console, connected through an adapter cable to a PC running the SDK version 1.0,' but it does not specify any details about the PC's hardware, such as CPU, GPU models, or memory. |
| Software Dependencies | Yes | The Kinect device used in our experiments was the 2010 model for the X-Box 360 video-game console, connected through an adapter cable to a PC running the SDK version 1.0. |
| Experiment Setup | Yes | KNN was set to K = 5, Manhattan distance as the distance metric and distance weighting of 1/d. For the MLP, we only considered networks with a single hidden layer and the number of hidden units was set to 40. Training was performed using the Backpropagation (Haykin 2008) algorithm with momentum set to 0.2, learning rate to 0.3 and 1000 maximum epochs. The SVM was trained using the Sequential Minimal Optimization (SMO) algorithm (Platt 1999), using a polynomial kernel and C = 100.0. |