A5039914830- Remote Sensing and LiDAR Applications
- Video Analysis and Summarization
- Remote-Sensing Image Classification
- Automated Road and Building Extraction
- Advanced Vision and Imaging
- Satellite Image Processing and Photogrammetry
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Anomaly Detection Techniques and Applications
- Face and Expression Recognition
- Human Pose and Action Recognition
- Biometric Identification and Security
- Sports Analytics and Performance
- Remote Sensing in Agriculture
- Face recognition and analysis
Purdue University West Lafayette
2015-2022
The SoccerNet 2022 challenges were the second annual video understanding organized by team. In 2022, composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving timestamps in long untrimmed videos, (2) replay grounding, live moment an shown a replay, (3) pitch localization, detecting line and goal part elements, (4) camera calibration, dedicated to intrinsic extrinsic parameters, (5) player re-identification, same players across multiple views, (6) object tracking, tracking...
In order to facilitate further research in stereo reconstruction with multi-date satellite images, the goal of this paper is provide a set stereo-rectified images and associated groundtruthed disparities for 10 AOIs (Area Interest) drawn from two sources: 8 IARPA's MVS Challenge dataset 2 CORE3D-Public dataset. The were by first constructing fused DSM pairs aligning 30 cm LiDAR DSM. Unlike existing benckmarking datasets, we have also carried out quantitative evaluation our using human...
Our goal here is threefold: [1] To present a new dense-stereo matching algorithm, tMGM, that by combining the hierarchical logic of tSGM with support structure MGM achieves 6-8\% performance improvement over baseline SGM (these numbers are posted under tMGM-16 in Middlebury Benchmark V3 ); and [2] Through an exhaustive quantitative qualitative comparative study, to compare how major variants approach dense stereo matching, including perform presence of: (a) illumination variations shadows,...
In order to construct algorithmic solutions the problem of geolocalization user-generated photographs and videos, one must first populate a geographic database with different types objects markings that are likely be seen in videos. Toward end, this paper presents framework for detecting labeling satellite imagery. The we interested characterized by low-level features exist mostly at or beyond limits spatial resolution images. To deal challenges posed extraction such from imagery, confine...
We present a novel multi-view training framework and CNN architecture for combining information from multiple overlapping satellite images noisy labels derived OpenStreetMap (OSM) to semantically label buildings roads across large geographic regions (100 km$^2$). Our approach semantic segmentation yields 4-7% improvement in the per-class IoU scores compared traditional approaches that use views independently of one another. A unique (and, perhaps, surprising) property our system is...
This work focuses on player re-identification in broadcast videos of team sports. Specifically, we focus identifying the same images captured from different camera viewpoints during any given moment a match. task differs traditional applications person re-id few important ways. Firstly, players wear highly similar clothes, thereby making it harder to tell them apart. Secondly, there are only number samples for each identity, which makes train system. Thirdly, resolutions often quite low and...