Data and Code

Model-Based Recovery of Fluid-Flow Parameters from Video
Included is video data and ground truth data of liquid film flowing over a spinning disk. Also includes a paper which analyzes the video data. The paper develops a novel approach for fluid tracking and analysis, and the results show that the speed of a disk and the flow rate are recovered with the error less than 10%.
More details
Video Analysis and Content Extraction (VACE)
Video Analysis and Content Extraction (VACE) was established to develop novel algorithms for automatic video content extraction, multi-modal fusion, and event understanding. In order to measure performance of these algorithms and gauge research progress, a series of evaluations focusing on 15 evaluation tasks were administered in this program. The program supports several evaluation conditions. Each evaluation condition was a combination of a technology class, target object type and a data domain. The technologies supported are: detection and tracking. The target object types supported are: face, hand, person, moving vehicle, text line and text word. The data domain supported are: broadcast new, meeting room, and street surveillance.
More details
Kinect Calibration Utility
Until recently, range imaging has been prohibitively expensive to implement due to equipment and computational requirements. Today, 3D data capture is almost a commodity with the release of Microsoft.s Kinect sensor: an inexpensive consumer device with an array of sensors, including a rigidly paired color camera and structured light range scanner. The range data and color image pair has previously relied on human input to determine homographic points during calibration, introducing potentially erroneous input. This software introduces a fully automated calibration procedure that requires no user input for homography, eliminating human error. Using this method, results yield a mean reprojection error of 0.867 pixels with a variance of 0.104, demonstrating superior accuracy over the device's built-in registration. The software requires OpenNI, OpenCV and .NET Framework 4.0 to be installed on the host PC.
More details

Detection of Thin Lines using Low-Quality Video from Low-Altitude Aircraft in Urban Settings
A novel thin line detection algorithm for use in low-altitude aerial vehicles is presented. This algorithm is able to detect thin obstacles such as cables, power lines, and wires. The system is intended to be used during urban search and rescue operations, capable of dealing with low-quality images, robust to image clutter, bad weather, and sensor artifacts. The detection process uses motion estimation at the pixel level, combined with edge detection, followed by a windowed Hough transform. The evidence of lines is tracked over time in the resulting parameter spaces using a dynamic line movement model. The algorithms receiver operating characteristic curve (ROC) is shown, based on a multi-site dataset with 86 videos with 10160 wires spanning in 5576 frames.
Download Dataset
Human ID Gait Challenge Dataset, Challenge Experiments, and Baseline
This is a large (1.2 Tera Byte) of video data of gait data from 122 in upto 32 possible combinations of variations in factors. Signature of release form is required. And, you have to send a hard drive with specifications mentioned in the website mentioned below.
Expected response time: 3 months.    Foreign distribution may be restricted.    More details
USF Human ID 3-D Database and Morphable Faces
Return the license agreement by physical mail (no FAX or emails) and we will mail CD with data. The license agreement MUST be signed by an entity authorized to make legal commitments on behalf of the institution employing the person. Read the terms of the license agreement carefully and make sure your institution agrees.
Expected response time: 3 months.    Foreign distribution may be restricted.    More details
Human Perception Based Edge Detector Evaluation
Code for Canny and Rothwell edge detectors are available here. Image data consists of 28 color images of man-made and natural objects, with and without texture. Edge data for 6 edge detectors are available for these images. These edge images were used in the human perception experiments.
More details
Min-cut, average-cut, and normalized-cut based grouping of edges
We have code to group straight-line edge segments based on perceptual organizational principles, using min-cut, average-cut, and normalized cut measures. We also have datasets of 50 and 100 images of man-made and natural objects, along with manually specified ground truth, to evaluate perceptual grouping algorithms.
More details
Perceptual Organization using Bayesian Networks
Code to detect perceptually significant geometric forms such as ellipses, circles, parallelograms, quandrilaterals, ribbons, and triangles and to track them through an image sequence. It uses graph theoretic operations and Bayesian networks to reason and hypothesize these structures.
More details
Sarkar-Boyer OZCO Edge Detector with Constant Curvature Edge Segmentation
Code to detect edges in an image, based on the Sarkar-Boyer edge detector. The code also includes an implementation of the real Canny edge detector. Edge detection is followed by edge linking, and segmentation of contours into straight lines and arc.
More details