Deborah Goshorn

Field Programmable Gate Array Implementation of Parts-Based Object Detection for Real Time Video Applications

The emergence of smart cameras has been fueled by increasingly advanced computing platforms that are capable of performing a variety of real-time computer vision algorithms. Smart cameras provide the ability to understand their environment. Object detection and behavior classification play an important role in making such observations. This paper presents a high-performance FPGA implementation of a generalized parts-based object detection and classifier that runs with capability of 266 frames/sec. The detection algorithm is easily reconfigured by simply loading a new representation into on-board memory, i.e., the FPGA can detect and classify a newly specified object and behavior without any changes to the hardware implementation.

Human posture recognition for intelligent vehicles

Pedestrian detection systems are finding their way into many modern “intelligent” vehicles. The body posture could reveal further insight about the pedestrian’s intent and her awareness of the oncoming car. This article details the algorithms and implementation of a library for real-time body posture recognition. It requires prior person detection and then calculates overall stance, torso orientation in four increments, and head location and orientation, all based on individual frames. A syntactic post-processing module takes temporal information into account and smoothes the results over time while correcting improbable configurations. We show accuracy and timing measurements for the library and its utilization in a training application.

Abnormal behavior-detection using sequential syntactical classification in a network of clustered cameras

Detecting abnormal behaviors is a critical task today. We need to monitor large areas, manage camera sensor data, and use this data for detecting behaviors, detecting the abnormal behaviors and classifying the normal behaviors. In order to monitor large areas, we need multiple cameras across a large-scale network. We use an architecture for a network of clustered cameras to minimize and efficiently manage bandwidth utilization. From this camera network architecture, we use the infrastructure outputs per cluster, per person, to detect abnormal behaviors intra-cluster; we also use the architecture outputs per person, per network, to detect global (inter-cluster) abnormal behaviors.

Behavior Modeling for Detection, Identification, Prediction, and Reaction (DIPR) in AI Systems Solutions

The application need for distributed artificial intelligence (AI) systems for behavior analysis and prediction is a requirement today versus a luxury of the past. The advent of distributed AI systems with large numbers of sensors and sensor types and unobtainable network bandwidth is also a key driving force. Additionally, the requirement to fuse a large number of sensor types and inputs is required and can now be implemented and automated in the AI hierarchy, and therefore, this will not require human power to observer, fuse, and interpret.

The Multi-level Learning and Classification of Multi-class Parts-Based Representations of U.S. Marine Postures

This paper primarily investigates the possibility of using multi-level learning of sparse parts-based representations of US Marine postures in an outside and often crowded environment for training exercises. To do so, the paper discusses two approaches to learning parts-based representations for each posture needed. The first approach uses a two-level learning method which consists of simple clustering of interest patches extracted from a set of training images for each posture, in addition to learning the nonparametric spatial frequency distribution of the clusters that represents one posture type. The second approach uses a two-level learning method which involves convolving interest patches with filters and in addition performing joint boosting on the spatial locations of the first level of learned parts in order to create a global set of parts that the various postures share in representation. Experimental results on video from actual US Marine training exercises are included.

Using a TI C6701 DSP Rapid Prototyping System for Nonlinear Adaptive Filtering to Mitigate Interference

Designing and implementing nonlinear systems, onto hardware devices experiences a paradigm shift with the innovative rapid prototyping system (RPS). The RPS is low cost, commercially off-the-shelf (COTS), and ingeniously establishes a direct path from initial design to a hardware implementation operating in real-time, eliminating several levels of tedious programming for hardware, with automatic code generation. In this paper, the hardware device is a digital signal processor (DSP) embedded in a prototype board. The RPS extends to real-world hardware testing specific to application, from which the nonlinear system design can be revised and optimized. From the RPS, final nonlinear system hardware can be designed with a high level of confidence, and the prototype board can continue to simulate other nonlinear devices. This paper explicates an overview of the RPS for nonlinear system hardware development. Section 2 addresses the nonlinear adaptive filter on the RPS for narrowband interference mitigation in communications channels. Section 3 conveys the overall RPS structure, procedure, and challenges overcome. Section 4 yields ideas for RPS expansion and concludes the paper.