Today, multi-sensor fusion detection frameworks in autonomous driving, especially sequence-based data-level frameworks, face high latency and coupling issues generally perform worse than LiDAR-only detectors. On this basis, we propose PMPF, point-cloud multiple-pixel fusion, for 3D object detection. PMPF projects the point cloud data onto image plane, where region pixels are processed to correspond with points decorated data, such that fused can be applied detectors autoencoders. is a...

In complex mining environments, driverless trucks are required to cooperate with multiple intelligent systems. They must perform obstacle avoidance based on factors such as the site road width, type, vehicle body movement state, and ground concavity-convexity. Targeting open-pit area, this paper proposes an object detection (IMOD) model developed using a 5G-multi-UAV deep learning approach. The IMOD employs data sensors monitor surface in real time within multisystem collaborative 5G...

Source-free domain adaptation in visual emotion recognition (SFDA-VER) is a highly challenging task that requires adapting VER models to the target without relying on source data, which of great significance for data privacy protection. However, due unignorable disparities between and traditional image classification existing SFDA methods perform poorly this task. In paper, we investigate SFDA-VER from fuzzy perspective identify two key issues: labels pseudo-labels. These issues arise...

Since the human faces are lowly textured, conventional stereo methods based on intensity correlation can not give satisfying 3D face reconstruction results. In this paper, a model matching method is proposed. A reference used as an intermedium for correspondence calculation. The virtual images with known correspondences first synthesized from face. Then extended to incoming images, using alignment and warping. complete thus be reconstructed reliably.

In this paper, we propose a novel method for eye states detection. The detection of is treated as an appearance based binary classification problem. whole region first scanned by series blocks with various locations and scales. Local Binary Pattern Histogram (LBPH) then extracted from each block to form descriptor the local texture. A reference template later calculated optimal histogram which makes distances between it LBPHs different clusters most separable. For all blocks, bin-wise...

In this paper, we propose a reliable method of eye states detection for drowsy driving monitoring. Given restricted local block regions, the Local Binary Pattern (LBP) histogram is extracted and each bin treated as feature eye. An AdaBoost based cascaded classifier then trained to select significant features from large sets classify open or closed. According eye, PERCLOS score measured in real time decide whether driver at state not. Experimental results demonstrate that our algorithm can...

Offline tracking of visual objects is particularly helpful in the presence significant occlusions, when a frame-by-frame, causal tracker likely to lose sight target. In addition, trajectories found by offline are typically smoother and more stable because global optimization this approach entails. contrast with previous work, we show that can be performed O(MNT) time for T frames video at M × N resolution, help generalized distance transform developed Felzenszwalb Huttenlocher [13]....

PDF HTML阅读 XML下载 导出引用 引用提醒 最小邻域均值投影函数及其在眼睛定位中的应用 DOI: 作者: 作者单位: 作者简介: 通讯作者: 中图分类号: 基金项目: Supported by the Opening Found of National Laboratory Pattern Recognition Institute Automation, Chinese Academy Sciences (中国科学院自动化研究所模式识别国家重点实验室开放基金); Graduate Innovation Fund USTC under Grant No.KD2006042 (中国科学技术大学研究生创新基金) Minimal Neighborhood Mean Projection Function and Its Application to Eye Location Author: Affiliation: Project: 摘要 | 图/表 访问统计 参考文献 相似文献 引证文献 资源附件 文章评论...

Many computer vision systems approximate targets' shape with rectangular bounding boxes. This choice trades localization accuracy for efficient computation. We propose twisted window search, a strict generalization over the globally optimal of target's shape. Despite its generality, we show that new algorithm runs in O(n <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> ), an asymptotic time complexity is no greater than search on image...

Target detection has a wide range of applications in many areas life, and it is also research hotspot the field unmanned driving. Urban roads are complex changeable, especially at intersections, which have always been difficult key part pilotless technology. Traffic policemen intersections link, but there few existing algorithms, speed generally slow. Aiming this problem, paper proposes real-time method traffic police based on YOLOv3 network.The YOLO network robust capable quickly completing...

In order to implement real‐time detection of passengers in subway stations, this paper proposes the SPDNet based on YOLOv4. Aiming at low accuracy station due uneven light conditions, we introduce attention mechanism CBAM recalibrate extracted features and improve robustness network. For crowded areas station, use K‐means++ algorithm generate anchors that are more consistent with passenger aspect ratio dataset KITTI, which mitigates missing caused by incorrect suppression true positive boxes...

Motion synchrony, i.e., the coordinated motion of a group individuals, is an interesting phenomenon in nature or daily life. Fish swim schools, birds fly flocks, soldiers march platoons, etc. Our goal to detect synchrony that may be present video data, and track moving objects as whole. This opens door novel algorithms applications. To this end, we model individual motions tubes space-time, define by geometric relation among tubes, whole set dynamic programming. The resulting algorithm...

In the pursuit of self-driving vehicles, pedestrian recognition plays an integral role. The following work proposes a gesture method based on k-nearest neighbour algorithm combined with pyramid residual module to reduce computation and improve real-time performance recognition. was formulated using data from Udacity, subsequently compared other methods. experimental results showed that accuracy new as high 92%, which is improvement over conventional histogram oriented gradient method....

For an automated driving system to be robust, it needs recognize not only fixed signals such as traffic signs and lights, but also gestures used by police. With the aim achieve this requirement, paper proposes a new gesture recognition technology based on graph convolutional network (GCN) according analysis of characteristics Chinese To begin, we spatial–temporal (ST-GCN) base while introducing attention mechanism, which enhanced effective features police balanced information distribution...

We present a nonparametric and efficient method for shape localization that improves on the traditional sub-window search in capturing fine geometry of an object from small number feature points. Our implies discrete set features capture more appearance information than is commonly exploited. use a-complex by Edelsbrunner et al. to build filtration simplicial complexes user-provided features. The optimal value determined automatically densest complex connected component, resulting...

This paper introduces a bilinear model to analyze and transfer expression or identity of 3D faces, its applications in 2D areas. Our aim is separate factors face data into two independent linear subspaces. First all the are proceeded have vertex-to-vertex correspondences. We build morphable transform these dense-vertex low-dimension parameter representation. Based on vectors denoting special we train identity-expression model, which spans training expressions identities by only few...

In this paper we propose a model-assisted binocular stereo algorithm for 3D face reconstruction. First, correspondence is reliably performed between input images by employing reference model as medium. Then, high-quality dataset of point cloud generated triangulation from the correspondence. Finally, an accurate and feature-preserving surface reconstructed denoising operation bilateral filtering meshing. We have compared faces with ground truth data. The experimental results show that...

Obstacle detection is one of the essential components navigation systems for autonomous vehicles. However, most previous studies have been directed at large obstacles; only a small number obstacles, which known as foreign object debris (FOD). In addition, these focused on specific scenarios, such airport runways. This paper addresses FOD conventional roads avoiding, decelerating driving, or having no effect driving. It proposes an improvement to MobileNet algorithm that streamlines network...

In the context of autonomous driving, effective recognition and understanding traffic police gestures is crucial for safe driving on road. Pedestrians' body movements affect system's gestures; therefore, localization key to gesture. With development deep learning technology, a satisfactory model requires huge amount accurately labelled datasets, but there are no publicly available relevant cost accurate labelling datasets extremely high, limiting performance model. To address this problem,...