Earticle

A safe and robust autonomous driving system relies on accurate perception of the environment for application-oriented scenarios. This paper proposes deployment of the three most crucial tasks (i.e., object detection, drivable area segmentation and lane detection tasks) on embedded system for self-driving operations. To achieve this research objective, multi-tasking network is utilized with a simple encoder-decoder architecture. Comprehensive and extensive comparisons for two models based on different backbone networks are performed. All training experiments are performed on server while Nvidia Jetson Xavier NX is chosen as deployment device.

This paper presents critical surveillance system functions and considers advances and challenges for privacy and ethical implications. We examine privacy protection strategies and responsible data management practices reflecting the critical problems of cybersecurity and data management to address concerns, ensuring that surveillance technology evolves ethically and sustainably. We highlight state-of-the-art deep learning-based techniques in object detection, addressing the complex difficulties of surveillance systems. Finally, we provide a research framework, including ethical and long-term implications and broad applications in privacy preservation and ethical responsibility. This study advances knowledge to ensure the long-term sustainability, security, and viability of information and communication technology applications.

Applications for object detection are expanding as it is automated through artificial intelligence-based processing, such as deep learning, on a large volume of images and videos. High dependence on training data and a non-transparent way to find answers are the common characteristics of deep learning. Attacks on training data and training models have emerged, which are closely related to the nature of deep learning. Privacy, integrity, and robustness for the extracted information are important security issues because deep learning enables object recognition in images and videos. This paper summarizes the security issues that need to be addressed for future applications and analyzes the state-of-the-art security studies related to robustness, privacy, and integrity of object detection for images and videos.

Three-dimensional (3D) object tracking is crucial in computer vision applications, particularly in autonomous driving, robotics, and surveillance. Despite advancements, effectively utilizing multimodal data to improve multi-object detection and tracking (MODT) remains challenging. This study introduces ACMODT, an affinity computation-based multi-object detection and tracking framework that integrates camera (2D) and LiDAR (3D) data for enhanced MODT performance in autonomous driving. This approach leverages EPNet as a backbone, utilizing 2D?3D feature fusion for accurate proposal generation. A deep neural network (DNN) extracts robust appearance and geometric features, while an improved affinity computation module combines Refined Boost Correlation Features (RBCF) and 3D-Extended Geometric IoU (3D-XGIoU) for precise object association. Motion prediction is refined using a Kalman filter (KF), and Gaussian Mixture Model (GMM)-based data association ensures consistent tracking. Experiments on the KITTI car tracking benchmark for quantitative analysis and the RADIATE dataset for visualization demonstrate that our method achieves superior tracking accuracy and precision compared to state-of-the-art multi-object tracking (MOT) approaches, proving its effectiveness for real-time object tracking.

In this paper, we produce Trash Object Detection (TOD) dataset to solve trash detection problems. A well-organized dataset of sufficient size is essential to train object detection models and apply them to specific tasks. However, existing trash datasets have only a few hundred images, which are not sufficient to train deep neural networks. Most datasets are classification datasets that simply classify categories without location information. In addition, existing datasets differ from the actual guidelines for separating and discharging recyclables because the category definition is primarily the shape of the object. To address these issues, we build and experiment with trash datasets larger than conventional trash datasets and have more than twice the resolution. It was intended for general household goods. And annotated based on guidelines for separating and discharging recyclables from the Ministry of Environment. Our dataset has 10 categories, and around 33K objects were annotated for around 5K images with 1280×720 resolution. The dataset, as well as the pre-trained models, have been released at https://github.com/jms0923/tod.

This paper introduces an edge AI-based scene-specific object detection system for long-term traffic management, focusing on analyzing congestion and movement via cameras. It aims to balance fast processing and accuracy in traffic flow data analysis using edge computing. We adapt the YOLOv5 model, with four heads, to a scene-specific model that utilizes the fixed camera's scene-specific properties. This model selectively detects objects based on scale by blocking nodes, ensuring only objects of certain sizes are identified. A decision module then selects the most suitable object detector for each scene, enhancing inference speed without significant accuracy loss, as demonstrated in our experiments.

Object detection is a fundamental yet challenging task in computer vision that plays an important role in object recognition, tracking, scene analysis and understanding. This paper aims to propose a multiproperty fusion algorithm for moving object detection. First, we build a scale-invariant feature transform (SIFT) vector field and analyze vectors in the SIFT vector field to divide vectors in the SIFT vector field into different classes. Second, the distance of each class is calculated by dispersion analysis. Next, the target and contour can be extracted, and then we segment the different images, reversal process and carry on morphological processing, the moving objects can be detected. The experimental results have good stability, accuracy and efficiency.

Small objects have become increasingly common in civil, industrial, and security sensitive environments, driving a strong demand for reliable vision-based detection and segmentation. While deep learning (DL) models have achieved impressive performance, their practicality is limited by the small sizes of objects, cluttered backgrounds, motion blur, domain shift, and the growing cost of training and inference. Quantum machine learning (QML) has recently emerged as a potential research direction for exploring alternative feature representations, optimization strategies, and learning paradigms under resource constrained conditions. However, its practical efficiency benefits for object detection and segmentation remain largely preliminary and context dependent. The literature on quantum methods for detection and segmentation is scattered across the QML, aerial vision, and remote sensing domains. We provide a structured review of quantum, hybrid quantum?classical (HQC), and quantum approaches for detection and segmentation. We classify existing methods based on four aspects: how features are encoded and represented in quantum systems, the use of quantum kernels, variational quantum circuits (VQC) for classification and detection, and quantum architectures combined with modern detection and segmentation models. We summarize the commonly used datasets for quantum techniques, evaluation protocols, and computational constraints, comparing the reported performance trends, scalability levels, and practical limitations of quantum-based methods. Finally, we identify the open challenges, including data-efficient learning, robustness to domain shift, and end-to-end quantum segmentation, along with the research directions relevant to bridging the gap between QML and real-world object perception systems.

The object classification using the images' contents is a big challenge in computer vision. The superpixels' information can be used to detect and classify objects in an image based on locations. In this paper, we proposed a methodology to detect and classify the image's pixels' locations using enhanced bag of words (BOW). It calculates the initial positions of each segment of an image using superpixels and then ranks it according to the region score. Further, this information is used to extract local and global features using a hybrid approach of Scale Invariant Feature Transform (SIFT) and GIST, respectively. To enhance the classification accuracy, the feature fusion technique is applied to combine local and global features vectors through weight parameter. The support vector machine classifier is a supervised algorithm is used for classification in order to analyze the proposed methodology. The Pascal Visual Object Classes Challenge 2007 (VOC2007) dataset is used in the experiment to test the results. The proposed approach gave the results in high-quality class for independent objects' locations with a mean average best overlap (MABO) of 0.833 at 1,500 locations resulting in a better detection rate. The results are compared with previous approaches and it is proved that it gave the better classification results for the non-rigid classes.

This paper presents a novel knock imaging technique for human-to-things interfaces. It allows any object to use its own surface as an imaging sensor. The methodology is to sense the structural bending waves caused by the knock and reconstruct the intensity image through the beamforming of the wave signals. To verify this, we prepare a thin piezoelectric sensor array on a square plate. The performances are evaluated by measuring the beam peak, beamwidth, and sidelobe level in the reconstructed image. Finally, a real finger knock test is demonstrated, showing an image reconstruction that the position error is only 3 mm.

In this study, we created a system that can determine the number of vehicles entering and departing a charging station in real time for solving waiting time problems during refueling. Accordingly, we use the You Only Look Once object detection algorithm to detect and count the number of vehicles in the charging station and send the data to the Firebase Realtime Database. The result is shown using an Android application that provides a map function with the Kakao Maps API at the user interface side. Our system has an accuracy of 91% and an average response time of 3.1 s. Therefore, this system can be used by drivers to determine the availability of a charging station and to identify the charging station with the least waiting time for charging their vehicle.

This paper presents a novel approach for identifying banknotes using an object detection algorithm to help tourists, particularly the elderly and visually impaired. Traditional banknote identification methods require specialized equipment. Thus, this work adopts an image-based method to provide a portable solution for enhancing convenience and accessibility. Furthermore, this paper compares two object detection algorithms, YOLOv8 and Faster R-CNN, to determine the optimal algorithm for banknote identification. The comparison results show that YOLOv8 achieves a mean average precision (mAP) of 0.973 at an intersection over union (IoU) threshold of 0.5, outperforming Faster R-CNN and previous methods by over 14%. The proposed system can identify the two most used currencies globally, the US Dollar and European Euro, demonstrating high accuracy and robustness in identifying multiple banknotes with mixed currencies within a single image. The trained model is implemented in a web application that is accessible everywhere, making the system portable.

Edges are a robust feature for object detection. In this paper, we present an edge-based background modeling method for the detection of moving objects. The edges in the image frames were mapped using robust Canny edge detector. Two edge maps were created and combined to calculate the ultimate moving-edge map. By selecting all the edge pixels of the current frame above the defined threshold of the ultimate moving edges, a temporary background-edge map was created. If the frequencies of the temporary background edge pixels for several frames were above the threshold, then those edge pixels were treated as background edge pixels. We conducted a performance comparison with previous works. The existing edge-based moving-object detection algorithms pose some difficulty due to the changes in background motion, object shape, illumination variation, and noises. The result of the performance evaluation shows that the proposed algorithm can detect moving objects efficiently in real-world scenarios.

This study proposes a mother-slave robot tracking system that identifies the target, predicts its location, and tracks it based on a single image. The proposed system utilizes a Convolutional Neural Network (CNN) for object detection, to identify the target robot. The distance and angle between the robots are then calculated through linear regression analysis, which offers a more efficient and cost-effective solution than traditional methods. The performance of the system was evaluated, resulting in an accuracy of 99.59% for object detection, and an average distance error of 2.04% for the estimated location.

With the growing popularity of smartphones, there has been an increase in the number of accidents involving users walking on stairs or crosswalks while using smartphones. Warning signs and images have been placed around dangerous locations in certain areas. However, this has not been significantly effective in reducing similar incidents. We propose a deep learning method based on object detection using a smartphone. Users are notified of impending detection risks on their smartphone’s screen. Tests demonstrated that our approach could detect stairs and crosswalks with high accuracy (96.7%). The proposed smartphone application includes deep learning network information, hyper-parameter information, and user-experience. Thus, users viewing their smartphone screens while walking can use the proposed solution to prevent accidents. As our knowledge, this is the first approach in the world to warn an imminent danger for smombies using a deep learning-based method.

In this paper, we expand the Phantom Sponge attack to monocular 3D object detection to increase false positives and detection times, thereby impairing the performance of edge devices. We introduce alpha-blending with a Universal Adversarial Patch (UAP) to modulate attack strength. Extensive experiments on KITTI, Rope3D, and NuScenes datasets across various hardware GPUs assess the UAP's impact. We also analyze power consumption, temperature, and other metrics on NVIDIA Jetson devices under UAP attacks. Our findings reveal the vulnerability of NMS-dependent models and edge devices to UAP attacks, the generalization of UAP, and the need for robust defenses in real-time applications.

It is important to understand the exact habitat distribution of endangered species because of their decreasing numbers. In this study, we build a system with a deep learning module that collects the image data of endangered animals, processes the data, and saves the data automatically. The system provides a more efficient way than human effort for classifying images and addresses two problems faced in previous studies. First, specious answers were suggested in those studies because the probability distributions of answer candidates were calculated even if the actual answer did not exist within the group. Second, when there were more than two entities in an image, only a single entity was focused on. We applied an object detection algorithm (YOLO) to resolve these problems. Our system has an average precision of 86.79%, a mean recall rate of 93.23%, and a processing speed of 13 frames per second.

Vision, Radar, and LiDAR sensors are widely used for autonomous vehicle perception technology. Especially object detection and classification are primarily dependent on vision sensors. However, under poor lighting conditions, dazzling sunlight, or bad weather an object might be difficult to be identified with general vision sensors. In this paper, we propose a sensor fusion system that combines a thermal infrared camera and a LiDAR sensor that can reliably detect and identify objects even in environments with poor visibility, such as day or night. The proposed method obtains the external parameters of the two sensors by designing and manufacturing a 3D calibration target to externally calibrate the thermal infrared camera and the LiDAR sensor. To verify the performance, experiments were conducted in day and night environments. The proposed sensor system and fusion algorithm show that it can reliably detect and identify objects even in environments with poor visibility, such as day or night.

Object fall detection is one of the significant applications in pose estimation. Traditional approaches heavily rely on fully mature neural networks, which may be complex to implement and resource-intensive. In this paper, we suggest a simple approach to detect Human fall on a single object using a mathematical comparison mechanism from 3D pose estimation landmarks lifted from 2D landmarks, suitable for low-specification systems. Our research focuses on dimensional lifting in 2D to 3D pose estimation based on object tracking-to adapt mapping 2D toons to 3D toons. For future research, we aim to develop a lightweight neural network for enhanced performance from ensemble effects and complex action detection. We also plan on enhancing the algorithm for multi-person detection. At this moment, our research extends into 3D toon generation. Our method achieves an F1-score of 94.7% (accuracy of 96.7%) with 28.2 FPS in detecting falls from webcam footage in a controlled environment.

With the mark of the fourth industrial revolution, the smart factory is evolving into a new future manufacturing plant. As a human-machine-interactive tool, augmented reality (AR) helps workers acquire the proficiency needed in smart factories. The valuable data displayed on the AR device must be delivered intuitively to users. Current AR applications used in smart factories lack user movement calibration, and visual fiducial markers for position correction are detected only nearby. This paper demonstrates a marker-based object detection using perspective projection to adjust augmented content while maintaining the user's original perspective with displacement. A new angle, location, and scaling values for the AR content can be calculated by comparing equivalent marker positions in two images. Two experiments were conducted to verify the implementation of the algorithm and its practicality in the smart factory. The markers were well-detected in both experiments, and the applicability in smart factories was verified by presenting appropriate displacement values for AR contents according to various movements.