- 标题
- 摘要
- 关键词
- 实验方案
- 产品
-
Getting to know low-light images with the Exclusively Dark dataset
摘要： Low-light is an inescapable element of our daily surroundings that greatly affects the efficiency of our vision. Research works on low-light imagery have seen a steady growth, particularly in the field of image enhancement, but there is still a lack of a go-to database as a benchmark. Besides, research fields that may assist us in low-light environments, such as object detection, has glossed over this aspect even though breakthroughs-after-breakthroughs had been achieved in recent years, most noticeably from the lack of low-light data (less than 2% of the total images) in successful public benchmark datasets such as PASCAL VOC, ImageNet, and Microsoft COCO. Thus, we propose the Exclusively Dark dataset to elevate this data drought. It consists exclusively of low-light images captured in visible light only, with image and object level annotations. Moreover, we share insightful findings in regards to the effects of low-light on the object detection task by analyzing the visualizations of both hand-crafted and learned features. We found that the effects of low-light reach far deeper into the features than can be solved by simple “illumination invariance”. It is our hope that this analysis and the Exclusively Dark dataset can encourage the growth in low-light domain researches on different fields. The dataset can be downloaded at https://github.com/cs-chan/Exclusively-Dark-Image-Dataset.
关键词： feature analysis,Low-light images,dataset,object detection,illumination invariance
更新于2025-09-10 09:29:36
-
Deformable Faster R-CNN with Aggregating Multi-Layer Features for Partially Occluded Object Detection in Optical Remote Sensing Images
摘要： The region-based convolutional networks have shown their remarkable ability for object detection in optical remote sensing images. However, the standard CNNs are inherently limited to model geometric transformations due to the fixed geometric structures in its building modules. To address this, we introduce a new module named deformable convolution that is integrated into the prevailing Faster R-CNN. By adding 2D offsets to the regular sampling grid in the standard convolution, it learns the augmenting spatial sampling locations in the modules from target tasks without additional supervision. In our work, a deformable Faster R-CNN is constructed by substituting the standard convolution layer with a deformable convolution layer in the last network stage. Besides, top-down and skip connections are adopted to produce a single high-level feature map of a fine resolution, on which the predictions are to be made. To make the model robust to occlusion, a simple yet effective data augmentation technique is proposed for training the convolutional neural network. Experimental results show that our deformable Faster R-CNN improves the mean average precision by a large margin on the SORSI and HRRS dataset.
关键词： Faster R-CNN,occluded object detection,data augmentation,Deformable CNN
更新于2025-09-10 09:29:36
-
[IEEE IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Valencia, Spain (2018.7.22-2018.7.27)] IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Object Detection for High-Resolution Sar Images Under the Spatial Constraints of Optical Images
摘要： With the rapid development of sensor technology, we pay more attention to object detection for high-resolution SAR images. Besides, the traditional object detection methods which only use one SAR image to accomplish detection aren’t enough appreciate for some cases that the background around the object is complex. In the paper, we propose an object detection method for high-resolution SAR images under the spatial constraints of optical images. It consists of three main steps: Establishment of the spatial relation between optical and SAR images, spatial constraints projection from optical images and detection under the spatial constraints in high-resolution SAR images. At the end of the paper, we take the inshore ship detection as an example to show that the proposed method can greatly improve the accuracy and efficiency of object detection against the traditional SAR object detection methods in conditions that background around the object is complex.
关键词： High-Resolution SAR images,Spatial Constraints,Complex Background,Object Detection
更新于2025-09-10 09:29:36
-
Traffic scene awareness for intelligent vehicles using ConvNets and stereo vision
摘要： In this paper, we propose an efficient approach to perform recognition and 3D localization of dynamic objects on images from a stereo camera, with the goal of gaining insight into traffic scenes in urban and road environments. We rely on a deep learning framework able to simultaneously identify a broad range of entities, such as vehicles, pedestrians or cyclists, with a frame rate compatible with the strict requirements of onboard automotive applications. Stereo information is later introduced to enrich the knowledge about the objects with geometrical information. The results demonstrate the capabilities of the perception system for a wide variety of situations, thus providing valuable information for a higher-level understanding of the traffic situation.
关键词： pose estimation,deep learning,object detection,intelligent vehicles
更新于2025-09-10 09:29:36
-
[IEEE IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Valencia, Spain (2018.7.22-2018.7.27)] IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Deconv R-CNN for Small Object Detection on Remote Sensing Images
摘要： Small object detection has drawn increasing interest in computer vision and remote sensing image processing. The Region Proposal Network (RPN) methods (e.g., Faster R-CNN) have obtained promising detection accuracy with several hundred proposals. However, due to the pooling layers in the network structure of the deep model, precise localization of small-size object is still a hard problem. In this paper, we design a network with a deconvolution layer after the last convolution layer of base network for small target detection. We call our model DeconvR-CNN. In the experiment on a remote sensing image dataset, DeconvR-CNN reaches a much higher mean average precision (mAP) than Faster R-CNN.
关键词： Object detection,Small object,Convolutional neural network,R-CNN,Deconvolution
更新于2025-09-09 09:28:46
-
[IEEE IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Valencia, Spain (2018.7.22-2018.7.27)] IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium - Virtualot — A Framework Enabling Real-Time Coordinate Transformation & Occlusion Sensitive Tracking Using UAS Products, Deep Learning Object Detection & Traditional Object Tracking Techniques
摘要： In this work we explore a combination of methods that allow us to analyze and study hyper-local environmental phenomena. Developing a unique application of monoplotting enables visualization of the results of deep-learning object detection and traditional object tracking processes applied to a perspective view of a parking lot on aerial imagery in real-time. Additionally, we propose a general algorithm to extract some scene understanding by inverting the monoplotting process and applying it to digital elevation models. This allows us to derive estimations of perspective image areas causing object occlusions. Connecting the real world and perspective spaces, we can create a resilient object tracking environment using both coordinate spaces to adapt tracking methods when objects encounter occlusions. We submit that this novel composite of techniques opens avenues for more intelligent, robust object tracking and detailed environment analysis using GIS in complex spatial domains provided video footage and UAS products.
关键词： computer vision,homography,object tracking,object detection,photogrammetry
更新于2025-09-09 09:28:46
-
Underwater Wide-Area Layered Light Field for Underwater Detection
摘要： In underwater electro-optic detection, image quality can be degraded by the backscattering of light from the illuminated water volume. In practical systems, we tend to simultaneously require a high level of detection distance (DD), ?eld of view (FOV), and depth of ?eld (DOF), but these factors in?uence each other by the media scattering. To eliminate this restriction, we propose to explore the underwater wide-area layered light ?eld (UWLLF), which classi?es the underwater detection area by the DD and distribution characteristics of the light ?eld, to minimize the scattering in?uence on target detection. Based on the UWLLF, an underwater electro-optic detection system is designed that can achieve the speci?cations of a 70? FOV and 7.9-fold attenuation length (for the attenuation coef?cient 1.43 /m of 532 nm) DD. In addition, with the spatial separation of light energy, the non-detection zone at short ranges is eliminated, yielding an almost full DOF. With these three factors simultaneously improved, the ability of underwater exploration for object detection is enhanced.
关键词： object detection,Underwater technology,light ?eld,optical imaging,underwater detection
更新于2025-09-09 09:28:46
-
[IEEE 2018 International Conference on 3D Vision (3DV) - Verona (2018.9.5-2018.9.8)] 2018 International Conference on 3D Vision (3DV) - SeeThrough: Finding Objects in Heavily Occluded Indoor Scene Images
摘要： Discovering 3D arrangements of objects from single indoor images is important given its many applications such as interior design and content creation for virtual environments. Although heavily researched in the recent years, existing approaches break down under medium to heavy occlusion as the core image-space region detection module fails in absence of directly visible cues. Instead, we take into account holistic contextual 3D information, exploiting the fact that objects in indoor scenes co-occur mostly in typical configurations. First, we use a neural network trained on real indoor annotated images to extract 2D keypoints, and feed them to a 3D candidate object generation stage. Then, we solve a global selection problem among these candidates using pairwise co-occurrence statistics discovered from a large 3D scene database. We iterate the process allowing for candidates with low keypoint response to be incrementally detected based on the location of the already discovered nearby objects. We demonstrate significant performance improvement over combinations of state-of-the-art methods, especially for scenes with moderately to severely occluded objects. Code and data available at http://geometry.cs.ucl.ac.uk/projects/2018/seethrough.
关键词： 3D vision,indoor scenes,object detection,occlusion,neural networks
更新于2025-09-09 09:28:46
-
[IEEE 2018 24th International Conference on Pattern Recognition (ICPR) - Beijing, China (2018.8.20-2018.8.24)] 2018 24th International Conference on Pattern Recognition (ICPR) - A Co-occurrence Background Model with Hypothesis on Degradation Modification for Object Detection in Strong Background Changes
摘要： Object detection has become an indispensable part of video processing and current background models are sensitive to background changes. In this paper, we propose a novel background model using an algorithm called Co-occurrence Pixel-block Pairs (CPB) against background changes, such as illumination changes and background motion. We utilize the co-occurrence “pixel to block” structure to extract the spatial-temporal information of each pixel to build background model, and then employ an efficient evaluation strategy to identify the current state of each pixel, which is named as correlation dependent decision function. Furthermore, we also introduce a Hypothesis on Degradation Modification (HoD) into CPB structure to reinforce the robustness of CPB. Experimental results obtained from the dataset of the PETS2001, AIST-Indoor, SBMnet and CDW-2012 databases show that our model can detect objects robustly in strong background changes.
关键词： co-occurrence pixel-block pairs,background model,Object detection,hypothesis on degradation modification
更新于2025-09-09 09:28:46
-
[IEEE 2018 24th International Conference on Pattern Recognition (ICPR) - Beijing, China (2018.8.20-2018.8.24)] 2018 24th International Conference on Pattern Recognition (ICPR) - Cross Modal Multiscale Fusion Net for Real-time RGB-D Detection
摘要： This paper presents a novel multi-modal CNN architecture for object detection by exploiting complementary input cues in addition to sole color information. Our one-stage architecture fuses the multiscale mid-level features from two individual feature extractor, so that our end-to-end net can accept crossmodal streams to obtain high-precision detection results. In comparison to other crossmodal fusion neural networks, our solution successfully reduces runtime to meet the real-time requirement with still high-level accuracy. Experimental evaluation on challenging NYUD2 dataset shows that our network achieves 49.1% mAP, and processes images in real-time at 35.3 frames per second on one single Nvidia GTX 1080 GPU. Compared to baseline one stage network SSD on RGB images which gets 39.2% mAP, our method has great accuracy improvement.
关键词： multi-modal CNN,fusion network,object detection,RGB-D,real-time
更新于2025-09-09 09:28:46