Non-maximum Suppression Research Articles

SFP-NMS: Nonmaximum Suppression for Suppressing False Positives During Merging Patches of High-Resolution Image

SFP-NMS: Nonmaximum Suppression for Suppressing False Positives During Merging Patches of High-Resolution Image

A DEEP LEARNING-BASED EFFICIENT FIREARMS MONITORING TECHNIQUE FOR BUILDING SECURE SMART CITIES

Any kind of violence is a shame to the world we live in. However, violence still plays a significant role in our society today and claims many innocent lives. Using a gun is one of the traditional methods of violence. Nowadays, firearm-related fatalities are a widespread occurrence. It poses a risk to public safety and presents difficulties for law enforcement. Most of these incidents take place in cities or semi-urban regions. Today, CCTV-based surveillance is widely used by both public and commercial institutions for monitoring and prevention. However, human-based monitoring is error-prone and needs a large number of person-hours as a resource. However, automated smart surveillance is better suited for large-scale and dependable monitoring of violent actions. The primary goal of the study is to demonstrate how deep learning-based methods may be used to identify weaponry, especially rifles. This research employs many detection methods for human face and gun identification, including the newest EfficientDet-based architectures and Faster Region-Based Convolutional Neural Networks (Faster RCNN). The post-processing detection performance of weapons and human faces has been enhanced using an ensemble (stacked) strategy that employs Non-Maximum Suppression, Non-Maximum Weighted, and Weighted Box Fusion techniques. In this study, the comparative outcomes of several detection methods and their ensembles have been addressed experimentally. It facilitates the police's rapid information gathering about the situation and enables them to take prompt preventative action. Additionally, social media videos may be found using the same method for gun-based content recognition. Here, the mean average precisions for the mAP0.5, mAP0.75, and mAP[0.500.95] are 77.02%, 16.40%, and 29.73%, respectively, according to the Weighted Box Fusion-based Ensemble Detection Scheme. Out of all the options tested, the outcomes provide the best performance. The model has undergone extensive testing using movie clips and unidentified test photos. The resulting ensemble schemes are consistently better than the basic models and are adequate.

A Method for Reducing False Negative Rate in Non-Maximum Suppression of YOLO Using Bounding Box Density

A Method for Reducing False Negative Rate in Non-Maximum Suppression of YOLO Using Bounding Box Density

Classification and Localization of Arabic Handwritten Text in KHATT Dataset Based on Faster R-CNN

Arabic handwriting recognition is an important research area in computer vision. Due to the complexity of the Arabic script, this is an arduous task. Several approaches have been proposed to address this challenge, including deep learning algorithms. Despite their efficiency, these approaches present some limitations such as the use of lexicon-driven models, the need of a lot of data for training and the huge computational cost. We propose, in this paper, two novel models based on the robust Faster Region-Convolution Neural Network (Faster R-CNN) to recognize Arabic handwritten sentences. The Faster R-CNN is commonly used to detect objects in images and depends on region proposal algorithms. These models use the pre-trained VGG 16 and ResNet50. Moreover, they use a soft non-maximum suppression strategy (Soft-NMS) in the post-processing stage instead of the traditional NMS to increase the detection rate of overlapping items. The models will be trained independently to identify words in the text lines and tested using sentences from the (KFUPM Handwritten Arabic TexT) KHATT dataset. It will be shown that the faster R-CNN models allow greater accuracy and effectiveness in handwriting recognition. They achieve accuracy rates of 99% and 98% for the two networks, VGG16 and ResNet50, respectively

Learning Visual Clue for UWB-based multi-person pose estimation

Compared to camera image-based methods, radio frequency (RF) based pose estimation has great potential for use in situations where the field of view is obstructed. In this paper, we present a novel RF-based Pose Estimation framework with Transformer (RPET) that operates in a fully end-to-end fashion and uses an easy-to-install portable radar. RPET eliminates the need for complex preprocessing and hand-crafted post-processing modules, such as region-of-interest (RoI) cropping, non-maximum suppression (NMS), and keypoint grouping. We also introduce a novel concept called Visual Clue (VC), which mimics a pose feature represented in image-based methods and improves the learning performance of multi-person pose estimation from RF signals. Our experimental results demonstrate the effectiveness of VC and the generalizability of our model to different environmental conditions, including changes in location and obstructed views.

Smart retail SKUs checkout using improved residual network

Intelligent signal processing in unmanned stores enhances operational efficiency, notably through automated SKUs (Stock Keeping Units) recognition, which expedites customer checkout. Distinguishing itself from generic detection algorithms, the retail product detection algorithm addresses challenges like densely arranged items, varying scales, large quantities, and product similarities. To mitigate these challenges, firstly we propose a novel boundary regression neural network architecture, which enhances the detection of bounding box in dense arrangement, minimizing computational costs and parameter sizes. Secondly, we propose a novel loss function for hierarchical detection, addressing imbalances in positive and negative samples. Thirdly, we enhance the conventional non-maximum suppression (NMS) with weighted non-maximum suppression (WNMS), tying NMS ranking scores to candidate box accuracy. Experimental results on SKU-110K and RPC datasets, two public available databases, show that the proposed SKUs recognition algorithm provides improved reliablity and efficiency over existing methods.

An Efficient Forest Fire Detection Algorithm Using Improved YOLOv5

Forest fires result in severe disaster, causing significant ecological damage and substantial economic losses. Flames and smoke represent the predominant characteristics of forest fires. However, these flames and smoke often exhibit irregular shapes, rendering them susceptible to erroneous positive or negative identifications, consequently compromising the overall performance of detection systems. To enhance the average precision and recall rates of detection, this paper introduces an enhanced iteration of the You Only Look Once version 5 (YOLOv5) algorithm. This advanced algorithm aims to achieve more effective fire detection. First, we use Switchable Atrous Convolution (SAC) in the backbone network of the traditional YOLOv5 to enhance the capture of a larger receptive field. Then, we introduce Polarized Self-Attention (PSA) to improve the modeling of long-range dependencies. Finally, we incorporate Soft Non-Maximum Suppression (Soft-NMS) to address issues related to missed detections and repeated detections of flames and smoke by the algorithm. Among the plethora of models explored, our proposed algorithm achieves a 2.0% improvement in mean Average Precision@0.5 (mAP50) and a 3.1% enhancement in Recall when compared with the YOLOv5 algorithm. The integration of SAC, PSA, and Soft-NMS significantly enhances the precision and efficiency of the detection algorithm. Moreover, the comprehensive algorithm proposed here can identify and detect key changes in various monitoring scenarios.

Multi-scale cross-layer fusion and center position network for pedestrian detection

Pedestrian detection has made breakthroughs after the rise of convolutional neural networks. However, it faces some challenging problems, including dataset difference, small pedestrian targets and occlusions between pedestrians. To deal with these problems, we propose a novel convolutional network architecture, named multi-scale cross-layer fusion and center position network (MCF-CP-NET). A new backbone unit is designed to introduce channel-wise attention into the improved aggregated residual transformations for effective feature extraction. We select suitable anchor setting for pedestrian detection datasets to tackle the problem of dataset difference. A feature pyramid sub-network with cross-layer fusion is developed for better detection of small pedestrian targets, where cross-layer connections are used to reduce the information loss and low-level marginal feature dissipation and better fuse low- and high-level features. We add a center position branch into the localization regression sub-network in MCF-CP-NET to better detect occluded pedestrians, which predicts the centrality index of the localization box to obtain the center score, and further optimizes the score of non-maximum suppression. Experiments show that the average precision and recall of MCF-CP-NET are separately improved by 1.2% and 0.7% on the person class of Pascal VOC2007 dataset and 1.6% and 0.1% on WiderPerson dataset, in comparison with the state-of-the-art.

High-accuracy low-latency non-maximum suppression processor for traffic object detection

High-accuracy low-latency non-maximum suppression processor for traffic object detection

Trusted multi-scale classification framework for whole slide image

Despite remarkable efforts, the classification of gigapixel whole-slide images (WSIs) is severely restricted by either limited computing resources or inadequate utilization of knowledge from different scales. Furthermore, previous attempts have often lacked the ability to estimate uncertainty for different scales. Typically, pathologists jointly analyze WSIs at different magnifications, changing the magnification repeatedly when uncertain to discover various tissue features. Motivated by this diagnostic process, we propose a trusted multi-scale classification framework for WSIs. Using the Vision Transformer as the backbone for multi-branches, our framework models joint classification, estimates uncertainty for each magnification of a microscope, and integrates evidence from different magnifications. Additionally, we propose a novel patch selection schema using attention rollout and non-maximum suppression to exploit discriminative patches from WSIs and reduce computational requirements. To empirically investigate the effectiveness of our approach, we conducted experiments on two benchmark databases for WSI classification. The results suggest that our trusted framework significantly improves WSI classification performance compared to state-of-the-art methods.

Sparse R-CNN: An End-to-End Framework for Object Detection.

Object detection serves as one of most fundamental computer vision tasks. Existing works on object detection heavily rely on dense object candidates, such as k anchor boxes pre-defined on all grids of an image feature map of size H×W. In this paper, we present Sparse R-CNN, a very simple and sparse method for object detection in images. In our method, a fixed sparse set of learned object proposals ( N in total) are provided to the object recognition head to perform classification and localization. By replacing HWk (up to hundreds of thousands) hand-designed object candidates with N (e.g., 100) learnable proposals, Sparse R-CNN makes all efforts related to object candidates design and one-to-many label assignment completely obsolete. More importantly, Sparse R-CNN directly outputs predictions without the non-maximum suppression (NMS) post-processing procedure. Thus, it establishes an end-to-end object detection framework. Sparse R-CNN demonstrates highly competitive accuracy, run-time and training convergence performance with the well-established detector baselines on the challenging COCO dataset and CrowdHuman dataset. We hope that our work can inspire re-thinking the convention of dense prior in object detectors and designing new high-performance detectors. Our code is available at: https://github.com/PeizeSun/SparseR-CNN.

Multi-scale split matching three-dimensional reconstruction method of surface microcracks on the silicon nitride bearing roller

Aiming at the characteristics of parametric discrete and non-significant depth of silicon nitride bearing roller surface microcracks. A multi-scale split matching method is proposed. To realize the full reconstruction of the characteristics of the three-dimensional morphology of the surface microcracks. Combined with the parametric discretization characteristics of the surface microcracks, a box filter is defined to do second-order Gaussian differentiation on the image. Construct a multiscale Hessian matrix in the scale space of the surface microcrack image. Determine the covariance-discrete feature points, and obtain the global covariance-discrete sparse point cloud of silicon nitride bearing roller surface microcracks by non-maximum value suppression. For the non-significant depth features of silicon nitride bearing roller surface microcracks. Dissect the surface microcrack distortion images. Divide each image into multiple regional pixel blocks and construct an aggregation function to determine the exact depth value of each pixel block at the cost of matching each pixel block with its surrounding pixel blocks. Dense point cloud reconstruction of non-significant depth features of silicon nitride bearing roller surface microcracks is realized. The average distribution of dense clusters of reprojection errors of the dense point cloud features is less than 0.4pix, the median residual of the field of view of the surface microcrack image is 0.57pix, and the total number of effectively extracted feature points is about 92.9 %. The detailed lossless recovery reconstruction of the parametric discrete, non-significant depth features of silicon nitride bearing roller surface microcracks is realized. A new idea is provided for the establishment of molecular dynamics geometric modeling of the surface microcrack extension process on the silicon nitride bearings roller.

Attention Mechanism and Detection Box Information Based Real-time Multi-Object Vehicle Detection

Ensuring both the accuracy of vehicle target detection and meeting real-time requirements is crucial in traffic videos. The YOLOv5s target detection framework, known for its accuracy and efficiency, has attracted attention in academic circles. However, there are still some features that can be optimized. First of all, the detection subnet of the YOLOv5s framework cannot smoothly convert complex feature maps into relatively sparse target prediction boxes. To solve this, we integrate a self-attention-based gating mechanism into the detection subnet, forming the YOLOv5s-SAG network. Secondly, the loss function of CIoU used by YOLOv5s pays insufficient attention to the overlapping area of the detection frame, which can be used as metric for measuring target detection effectiveness. We add the loss term of area ratio to CIoU to further improve the modeling ability. Finally, the current multi-class Non-Maximum Suppression algorithm can cause high overlap of multi-class detection frames. To improve it, we propose a multi-class CS-NMS algorithm based on category suppression. Experimental results show an approximately 8% improvement in the mAP50 index on the UA-DETRAC dataset compared with YOLOv5s. The proposed algorithm also achieves better detection results compared to mainstream target detection algorithms and meets the real-time requirements of traffic video analysis.

Joint Object Detection and Re-Identification for 3D Obstacle Multi-Camera Systems

The growing on-board processing capabilities have led to more complex sensor configurations, enabling autonomous car prototypes to expand their operational scope. Nowadays, the joint use of LiDAR data and multiple cameras is almost a standard and poses new challenges for existing multi-modal perception pipelines, such as dealing with contradictory or redundant detections caused by inference on overlapping images. In this paper, we address this last issue in the context of sequential schemes like F-PointNets, where object candidates are obtained in the image space, and the final 3D bounding box is then inferred from point cloud information. To this end, we propose the inclusion of a re-identification branch into the 2D detector, i.e., Faster R-CNN, so that objects seen from adjacent cameras can be handled before the 3D box estimation takes place, removing duplicates and completing the object’s cloud. Extensive experimental evaluations covering both the 2D and 3D domains affirm the effectiveness of the suggested methodology. The findings indicate that our approach outperforms conventional Non-Maximum Suppression (NMS) methods. Particularly, we observed a significant gain of over 5% in terms of accuracy for cars in camera overlap regions. These results highlight the potential of our upgraded detection and re-identification system in practical scenarios for autonomous driving.

Enhancing Tuta absoluta Detection on Tomato Plants: Ensemble Techniques and Deep Learning

Early detection and efficient management practices to control Tuta absoluta (Meyrick) infestation is crucial for safeguarding tomato production yield and minimizing economic losses. This study investigates the detection of T. absoluta infestation on tomato plants using object detection models combined with ensemble techniques. Additionally, this study highlights the importance of utilizing a dataset captured in real settings in open-field and greenhouse environments to address the complexity of real-life challenges in object detection of plant health scenarios. The effectiveness of deep-learning-based models, including Faster R-CNN and RetinaNet, was evaluated in terms of detecting T. absoluta damage. The initial model evaluations revealed diminishing performance levels across various model configurations, including different backbones and heads. To enhance detection predictions and improve mean Average Precision (mAP) scores, ensemble techniques were applied such as Non-Maximum Suppression (NMS), Soft Non-Maximum Suppression (Soft NMS), Non-Maximum Weighted (NMW), and Weighted Boxes Fusion (WBF). The outcomes shown that the WBF technique significantly improved the mAP scores, resulting in a 20% improvement from 0.58 (max mAP from individual models) to 0.70. The results of this study contribute to the field of agricultural pest detection by emphasizing the potential of deep learning and ensemble techniques in improving the accuracy and reliability of object detection models.

End-to-End Joint Multi-Object Detection and Tracking for Intelligent Transportation Systems

Environment perception is one of the most critical technology of intelligent transportation systems (ITS). Motion interaction between multiple vehicles in ITS makes it important to perform multi-object tracking (MOT). However, most existing MOT algorithms follow the tracking-by-detection framework, which separates detection and tracking into two independent segments and limit the global efficiency. Recently, a few algorithms have combined feature extraction into one network; however, the tracking portion continues to rely on data association, and requires complex post-processing for life cycle management. Those methods do not combine detection and tracking efficiently. This paper presents a novel network to realize joint multi-object detection and tracking in an end-to-end manner for ITS, named as global correlation network (GCNet). Unlike most object detection methods, GCNet introduces a global correlation layer for regression of absolute size and coordinates of bounding boxes, instead of offsetting predictions. The pipeline of detection and tracking in GCNet is conceptually simple, and does not require complicated tracking strategies such as non-maximum suppression and data association. GCNet was evaluated on a multi-vehicle tracking dataset, UA-DETRAC, demonstrating promising performance compared to state-of-the-art detectors and trackers.

Improvement of Road Instance Segmentation Algorithm Based on the Modified Mask R-CNN

Although the Mask region-based convolutional neural network (R-CNN) model possessed a dominant position for complex and variable road scene segmentation, some problems still existed, including insufficient feature expressive ability and low segmentation accuracy. To address these problems, a novel road scene segmentation algorithm based on the modified Mask R-CNN was proposed. The multi-scale backbone network, Res2Net, was utilized to replace the ResNet network, and aimed to improve the feature extraction capability. The soft non-maximum suppression algorithm with attenuation function (soft-NMS) was adopted to improve detection efficiency in the case of a higher overlap rate. The comparison analyses of partition accuracy for various models were performed on the adopted Cityscapes dataset. The results demonstrated that the modified Mask R-CNN effectively increased the segmentation accuracy, especially for small and highly overlapping objects. The adopted Res2Net and soft-NMS can effectively enhance the feature extraction and improve segmentation performance. The average accuracy of the modified Mask R-CNN model reached up to 0.321, and was 0.054 higher than Mask R-CNN. This work provides important guidance to design a more efficient road scene instance segmentation algorithm for further promoting the actual application in automatic driving systems.

Optimized faster R-CNN for oil wells detection from high-resolution remote sensing images

ABSTRACTAs the oil and gas industry is crucial to the global energy market, policymakers need accurate information about local oil reserves and the harmful environmental effects of drilling, such as damage to public land and wildlife. Therefore, accurate automatic detection of the location, distribution and quantity of oil wells is essential. Recent advancements in remote sensing and deep learning technologies provide potential solutions for automatic oil wells detection using high-resolution remote sensing images. This study proposes an optimized Faster R-CNN-based model that incorporates three key modifications to improve the accuracy of oil wells detection. The modifications include replacing the VGG16 network with the ResNet50 network to improve the model’s feature extraction capabilities, substituting the ordinary convolution of ResNet with a dilated convolution to improve the model’s receptive field, and constructing a feature pyramid to improve the model’s ability to detect small targets and objects at different scales. Also, an edge detection module is added to further improve the detection accuracy. Furthermore, a new framework based on Faster R-CNN and leveraging Soft-NMS (Non-Maximum Suppression) and the proposed ClusterRPN sub-network is combined to address the problem of clustered oil wells detection. Experimental results demonstrate that the proposed optimized model outperforms existing models.

Real-Time Object Detection Overview: Advancements, Challenges, and Applications

Real-time object detection is a crucial aspect of computer vision with applications spanning autonomous vehicles, surveillance, robotics, and augmented reality. This study examines real-time object detection techniques, highlighting their significance in artificial intelligence. The primary goal is swift and accurate object identification in images or video streams. Traditional methods like sliding windows and region-based approaches had limitations in computational efficiency. Deep learning, particularly Convolutional Neural Networks (CNNs), revolutionized object detection. Models like SSD, YOLO, and Faster R-CNN excel in accuracy and speed. They employ anchor boxes, feature pyramid networks, and non-maximum suppression to balance precision and processing speed. Hardware accelerators like GPUs, TPUs, and FPGAs facilitate real-time inference.
 Challenges in real-time object detection include occlusion, scale variations, and cluttered environments. Researchers must navigate the trade-offs between accuracy and speed. Real-time object detection is pivotal in computer vision, enabling intelligent systems across diverse applications. The continuous evolution of deep learning algorithms and hardware capabilities pushes the boundaries of this field, making it a dynamic research domain in artificial intelligence.

Dense-and-Similar Object detection in aerial images

The general object detection performance has been improving significantly due to the prosperity of deep learning. When applied to aerial images, these algorithms perform poorly. There are, as we summarized, two practical reasons: (1) photographed from far above, objects are commonly small in size, dense in clusters, but sparsely scattered in the whole image; (2) some classes are naturally similar in appearance, not to mention the insufficiency of distinguishable visual features captured by cameras due to factors such as bad weather, various angles of view and long distances, etc. To address the two issues, we propose a dense-and-similar object detector with four key components: a coarse detector, an adaptive clustering procedure, a similar-class classifier, and a fine detector. At first, we input the original image into both the coarse and fine detectors. Then we cluster and patch the coarsely detected results adaptively to form a foreground region image. We feed the patched image into the coarse detector again and use the similar-class classifier to re-identify the labels of detected bounding boxes. At last, we put re-identified detections and outputs of the fine detector together, and obtain the final detections after non-maximal suppression. The approach proposed in this study achieved excellent results on both the VisDrone 2019 and DIOR datasets through experimental validation.