Olga Mendoza-Schrock

An end-to-end vechicle classification pipeline using vibrometry data

This paper evaluates and expands upon the existing end-to-end process used for vibrometry target classification and identification. A fundamental challenge in vehicle classification using vibrometry signature data is the determination of robust signal features. The methodology used in this paper involves comparing the performance of features taken from automatic speech recognition, seismology, and structural analysis work. These features provide a means to reduce the dimensionality of the data for the possibility of improved separability. The performances of different groups of features are compared to determine the best feature set for vehicle classification. Standard performance metrics are implemented to provide a method of evaluation. The contribution of this paper is to (1) thoroughly explain the time domain and frequency domain features that have been recently applied to the vehicle classification using laser-vibrometry data domain, (2) build an end-to-end classification pipeline for Aided Target Recognition (ATR) with common and easily accessible tools, and (3) apply feature selection methods to the end-to-end pipeline. The end-to-end process used here provides a structured path for accomplishing vibrometry-based target identification. This paper will compare with two studies in the public domain. The techniques utilized in this paper were utilized to analyze a small in-house database of several different vehicles.

Applying Manifold Learning to Vehicle Classification Using Vibrometry Signatures

Understanding and organizing data is the first step toward exploiting laser vibrometry sensor phenomenology for target classification. A fundamental challenge in robust vehicle classification using vibrometry signature data is the determination of salient signal features and the fusion of appropriate measurements. . A particular technique, Diffusion Maps, has demonstrated the potential to extract intuitively meaningful features [1]. We want to develop an understanding of this technique by validating existing results using vibrometry data. This paper briefly describes the Diffusion Map technique, its application to dimension reduction of vibrometry data, and describes interesting problems to be further explored.

Electro-optical synthetic civilian vehicle data domes

This paper will look at using open source tools (Blender, LuxRender, and Python) to generate a large data set to be used to train an object recognition system. The model produces camera position, camera attitude, and synthetic camera data that can be used for exploitation purposes. We focus on electro-optical (EO) visible sensors to simplify the rendering but this work could be extended to use other rendering tools that support different modalities. The key idea of this paper is to provide an architecture to produce synthetic training data which is modular in design and constructed on open-source off-the-shelf software yielding a physics accurate virtual model of the object we want to recognize. For this paper the objects we are focused on are civilian vehicles. This architecture shows how leveraging existing open-source software allows for practical training of Electro-Optical object recognition algorithms.

Applying Manifold Learning Techniques to the CAESAR Database

Understanding and organizing data is the first step toward exploiting sensor phenomenology for dismount tracking. What image features are good for distinguishing people and what measurements, or combination of measurements, can be used to classify the dataset by demographics including gender, age, and race? A particular technique, Diffusion Maps, has demonstrated the potential to extract features that intuitively make sense [1]. We want to develop an understanding of this tool by validating existing results on the Civilian American and European Surface Anthropometry Resource (CAESAR) database. This database, provided by the Air Force Research Laboratory (AFRL) Human Effectiveness Directorate and SAE International, is a rich dataset which includes 40 traditional, anthropometric measurements of 4400 human subjects. If we could specifically measure the defining features for classification, from this database, then the future question will then be to determine a subset of these features that can be measured from imagery. This paper briefly describes the Diffusion Map technique, shows potential for dimension reduction of the CAESAR database, and describes interesting problems to be further explored.

Exploring Image Registration Techniques for Layered Sensing

In this paper we evaluate several methods to register and stabilize a motion imagery video sequence under the layered sensing construction. Layered sensing is a new construct in the repertoire of the US Air Force. Under the layered sensing paradigm, an area is surveyed by a multitude of sensors at many different altitudes and operating across many modalities. This combination of sensors provides better insight into a situation than could ever be achieved with a single sensor. A fundamental requirement to utilize this technology is to first register and stabilize the data from each of the individual sensors. The contribution of this paper is to explore and provide a preliminary evaluation of techniques for image registration of Electro-Optical (EO) video sequences taken from Wide Area Persistent Surveillance (WAPS) platforms whose views are centered on a city. Additionally, evaluation metrics for such techniques are described and explored.

Diffusion maps for exploring electro-optical synthetic vehicle image data

In this work, we explore low-dimensional representations of high-dimensional data derived from electro-optical synthetic vehicle images. The collection of vehicle images consists of four different vehicle models: Toyota Camry, Toyota Avalon, Toyota Tacoma, and Nissan Sentra. This data contains 3,601 160 × 213 gray-scale vehicle images sampled uniformly over a camera view hemisphere. We use the non-linear manifold learning technique of diffusion maps with Gaussian kernel to explore low-dimensional structure the high-dimensional cloud of vehicle image observations. Diffusion maps have been shown to be a valuable tool in the analysis of high-dimensional data and the technique is able to extract an approximation for the underlying structure inherent to the data. Our analysis includes examining how the diffusion time and kernel width leads to different low-dimensional representations and we present a novel technique to relate the kernel width to the distribution of data in the observation space. In addition, we present initial results for multi-class vehicle classification through low-dimensional embedding coordinates and the out-of-sample extension of unlabeled vehicle images.

Evolving robust gender classification features for CAESAR data

In this paper we explore the robustness of histogram features extracted from 3D point clouds of human subjects for gender classification. Experiments are conducted using point clouds drawn from the Civilian American and European Surface Anthropometry Resource Project (CAESAR anthropometric database provided by the Air Force Research Laboratory (AFRL) Human Effectiveness Directorate and SAE International). This database contains approximately 4400 high resolution LIDAR whole body scans of carefully posed human subjects. Features are extracted from each point cloud by embedding the cloud in series of cylindrical shapes and computing a point count for each cylinder that characterizes a region of the subject. These measurements define rotationally invariant histogram features that are processed by a classifier to label the gender of each subject. Preliminary results using cylinder sizes defined by human experts demonstrate that gender can be predicted with 98% accuracy for the type of high density point cloud found in the CAESAR database. In our previous has shown that when point cloud densities are reduced to levels that might be obtained using stand-off sensors; gender classification accuracy degrades. In this paper we show the results of how the classification accuracy degrades as a function of center of mass displacements.

A scaled, performance driven evaluation of the layered-sensing framework utilizing polarimetric infrared imagery

The layered sensing framework, in application, provides a useful, but complex integration of information sources, e.g. multiple sensing modalities and operating conditions. It is the implied trade-off between sensor fidelity and system complexity that we address here. Abstractly, each sensor/source of information in a layered sensing application can be viewed as a node in the network of constituent sensors. Regardless of the sensing modality, location, scope, etc., each sensor collects information locally to be utilized by the system as a whole for further exploitation. Consequently, the information may be distributed throughout the network and not necessarily coalesced in a central node/location. We present, initially, an analysis of polarimetric infrared data, with two novel features, as one of the input modalities to such a system. We then proceed with statistical and geometric analyses of an example network, thus quantifying the advantages and drawbacks of a specific application of the layered sensing framework.

Evolving point-cloud features for gender classification

In this paper we explore the use of histogram features extracted from 3D point clouds of human subjects for gender classification. Experiments are conducted using point clouds drawn from the CAESAR anthropometric database provided by the Air Force Research Laboratory (AFRL) Human Effectiveness Directorate and SAE International. This database contains approximately 4400 high resolution LIDAR whole body scans of carefully posed human subjects. Features are extracted from each point cloud by embedding the cloud in series of cylindrical shapes and computing a point count for each cylinder that characterizes a region of the subject. These measurements define rotationally invariant histogram features that are processed by a classifier to label the gender of each subject. Preliminary results using cylinder sizes defined by human experts demonstrate that gender can be predicted with 98% accuracy for the type of high density point cloud found in the CAESAR database. When point cloud densities are reduced to levels that might be obtained using stand-off sensors; gender classification accuracy degrades. We introduce an evolutionary algorithm to optimize the number and size of the cylinders used to define histogram features. The objective of this optimization process is to identify a set of cylindrical features that reduces the error rate when predicting gender from low density point clouds. A wrapper approach is used to interleave feature selection with classifier evaluation to train the evolutionary algorithm. Results of classification accuracy achieved using the evolved features are compared to the baseline feature set defined by human experts.

A limited comparative study of dimension reduction techniques on CAESAR

Understanding and organizing data is the first step toward exploiting sensor phenomenology. What features are good for distinguishing people and what measurements, or combination of measurements, can be used to classify people by demographic characteristics including gender? Dimension reduction techniques such as Diffusion Maps that intuitively make sense [1] and Principal Component Analysis (PCA) have demonstrated the potential to aid in extracting such features. This paper briefly describes the Diffusion Map technique and PCA. More importantly, it compares two different classifiers, K-Nearest Neighbors (KNN) and Adaptive boost (Adaboost), for gender classification using these two dimension reduction techniques. The results are compared on the Civilian American and European Surface Anthropometry Resource Project (CAESAR) database, provided by the Air Force Research Laboratory (AFRL) Human Effectiveness Directorate and SAE International. We also compare the results described herein with those of other classification work performed on the same dataset, for completeness.