A5053700968- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Video Surveillance and Tracking Methods
- Image and Object Detection Techniques
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Computer Graphics and Visualization Techniques
- Remote Sensing and LiDAR Applications
- Vehicle License Plate Recognition
- Augmented Reality Applications
- Image Retrieval and Classification Techniques
- 3D Shape Modeling and Analysis
- Optical measurement and interference techniques
- Video Analysis and Summarization
- Image Processing and 3D Reconstruction
- 3D Surveying and Cultural Heritage
- Autonomous Vehicle Technology and Safety
- Face and Expression Recognition
- Medical Imaging and Analysis
- Craniofacial Disorders and Treatments
- Remote-Sensing Image Classification
- Handwritten Text Recognition Techniques
- Indoor and Outdoor Localization Technologies
- Human Pose and Action Recognition
- QR Code Applications and Technologies
Brno University of Technology
2014-2023
University of Technology
2015
We are dealing with the problem of fine-grained vehicle make&model recognition and verification. Our contribution is showing that extracting additional data from video stream - besides image itself feeding it into deep convolutional neural network boosts performance considerably. This information includes: 3D bounding box used for "unpacking" image, its rasterized low-resolution shape, about orientation. Experiments show adding such decreases classification error by 26% (the accuracy...
This paper proposes an approach to the vehicle reidentification problem in a multiple camera system. We focused on re-identification itself assuming that detection is already solved including extraction of full-fledged 3D bounding box. The by using color histograms and oriented gradients linear regressor. features are used separate models order get best results shortest CPU computation time. proposed method works with high accuracy (60% true positives retrieved 10% false positive rate...
In this paper, we focus on fine-grained recognition of vehicles mainly in traffic surveillance applications. We propose an approach that is orthogonal to recent advancements (automatic part discovery and bilinear pooling). addition, contrast other methods focused vehicles, do not limit ourselves a frontal/rear viewpoint, but allow the be seen from any viewpoint. Our based 3-D bounding boxes built around vehicles. The box can automatically constructed data. For scenarios where it possible use...
This work is focused on recognition of license plates in low resolution and quality images. We present a methodology for collection real world (non-synthetic) dataset plate images with ground truth transcriptions. Our approach to the based Convolutional Neural Network which holistically processes whole image, avoiding segmentation characters. Evaluation results multiple datasets show that our method significantly outperforms other free commercial solutions data. To enable further research...
This paper deals with automatic calibration of roadside surveillance cameras. We focus on parameters necessary for measurements in traffic-surveillance applications. Contrary to the existing solutions, our approach requires no a priori knowledge, and it works very wide variety road settings (number lanes, occlusion, quality ground marking), as well practically unlimited viewing angles. The main contribution is that solution fully automatically-without any percamera or per-video manual input...
We present a novel way of odometry estimation from Velodyne LiDAR point cloud scans. The aim our work is to overcome the most painful issues data - sparsity and quantity points in an efficient way, enabling more precise registration. Alignment clouds which yields final based on random sampling using Collar Line Segments (CLS). closest line segment pairs are identified two sets segments obtained consequent From each pair correspondences, transformation aligning matched into 3D plane...
We introduce a novel method for odometry estimation using convolutional neural networks from 3D LiDAR scans. The original sparse data are encoded into 2D matrices the training of proposed and prediction. Our show significantly better precision in translational motion parameters comparing with state art LOAM, while achieving real-time performance. Together IMU support, high quality registration is realized. Moreover, we propose alternative CNNs trained prediction rotational results also...
In this paper, we focus on traffic camera calibration and a visual speed measurement from single monocular camera, which is an important task of surveillance. Existing methods addressing problem are difficult to compare due lack common data set with reliable ground truth. Therefore, it not clear how the in various aspects what factors affecting their performance. We captured new 18 full-HD videos, each around 1 hr long, at six different locations. Vehicles videos (20865 instances total)...
This paper deals with detection and recognition of matrix codes, such as the QR in high-resolution images real-world scenes. The goal is to provide a detector capable operation real time even on (several megapixels). We present an efficient algorithm for possible occurrences codes. characterized by very low false negative rate reasonable alarm rate. results our are be followed accurate detection/recognition algorithm. propose use recent code based Hough transform, because it can reuse some...
This paper presents a novel method for ground segmentation in Velodyne point clouds. We propose an encoding of sparse 3D data from the sensor suitable training convolutional neural network (CNN). general purpose approach is used cloud into and non-ground points. The LiDAR are represented as multi-channel 2D signal where horizontal axis corresponds to rotation angle vertical represents channels - laser beams. Multiple topologies relatively shallow CNNs (i.e. 3-5 layers) trained evaluated,...
The 1 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">st</sup> Workshop on Maritime Computer Vision (MaCVi) 2023 focused maritime computer vision for Unmanned Aerial Vehicles (UAV) and Surface Vehicle (USV), organized several subchallenges in this domain: (i) UAV-based Object Detection, (ii) Mar-itime Tracking, (iii) USV-based Obstacle Segmentation (iv) Detection. were based the SeaDronesSee MODS benchmarks. This report summarizes main findings...
Detection of lines in raster images is often performed using Hough transform. This paper presents a new parameterization and modification the transform–PClines. PClines are based on parallel coordinates, coordinate system used mostly or solely for high-dimensional data visualization. The algorithm described paper; its accuracy evaluated numerically compared to commonly line detectors results show that outperform existing approaches terms accuracy. Besides, computationally extremely...
Ray-triangle intersection is an important algorithm, not only in the field of realistic rendering (based on ray tracing) but also physics simulation, collision detection, modeling, etc. Obviously, speed this well-defined algorithm's implementations because calls to such a routine are numerous and simulation applications. Contemporary fast algorithms, which use SIMD instructions, focus packets against triangles. For between single rays triangles, operations as horizontal addition or dot...
Physics-based approaches could simplify terrain modeling by increasing its realism. However, most simulations provide only a low level of user control because they fail on large-scale phenomena or focus the limited effects. A new physics-based system for digital editing is suitable digital-content authors such as game designers, artists, and 3D modelers. It doesn't assume in-depth knowledge about simulations. Users can load large terrains from external sources, generate them procedurally,...
We present an object detector coupled with pose estimation directly in a single compact and simple model, where the shares extracted image features estimator. The output of classification each candidate window consists both score likelihood map poses. This extension introduces negligible overhead during detection so that is still capable real time operation. evaluated proposed approach on problem vehicle detection. used existing datasets viewpoint/pose annotation (WCVP, 3D objects, KITTI)....