Figure 4 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
Counting objects in video images has been an active area of computer vision for decades. For precise counting, it is necessary to detect objects and follow them through consecutive frames. Deep neural networks have allowed great improvements in this area. Nonetheless, this task is still a challenge for edge computing, especially when low-power edge...
Context in source publication
Context 1
... back to the last test group results, the biggest problem encountered was the very high error rate when counting people in the direction from right to left. The first step to know the origin of this problem was to observe the confusion matrix of the people counter in this same direction (Figure 4). It is clear that the main source of the problem was the system counting people when it was not supposed to. ...
Citations
... The tagging provided by unary all regression alone is typically noisy and unreliable [5]. YOLO is a well-known deep learning model for object identification and counting, and its model has been demonstrated to reach state-of-the-art performance on a wide range of benchmark datasets, all while maintaining high accuracy and smooth frame rates on high-end GPUs [7] [8]. ...
The ability to accurately recognize and count persons is crucial in many real-world applications, including surveillance, security, and crowd management, making it one of computer vision’s most fundamental tasks. You Only Look Once (YOLO) is one of the most effective deep learning models for object identification and counting in recent years. This research seeks to learn more about the YOLOv8 algorithm for precisely counting people in still photos and moving videos. The YOLO method has been at the forefront of computer vision due to its ability to recognize things in real time. People in a crowd typically overlap and block one other, and perspective effects can result in enormous changes in human size, shape, and appearance in the image, all of which make accurate headcounts challenging.The YOLO methodology and its adaptation for population census are the subject of this research. Results from experiments support the usefulness of the proposed approach. Surveillance, crowd control, traffic monitoring, retail analytics, event management, and urban planning are just some of the potential uses highlighted by the findings of this study. Mean Average Precision (MAP) numbers demonstrate that the identification procedure was successful, and the counting process was accurate to within 100%.
... In [24], a system was developed for real-time counting of people and bicycles using a private dataset of 339 videos. The dataset consisted of 4 s videos showing people and bicycles in motion, from which 630 frames were extracted and annotated on the Roboflow platform. ...
... Tensorflow, TensorRT, Keras, and PyTorch are examples of ML and deep learning frameworks. In the article [23], TensorRT was used and in [24], a comparison of accuracy between TensorRT and PyTorch was carried out. Tensorflow was the framework chosen in [18,20]. ...
Hiking and cycling have become popular activities for promoting well-being and physical activity. Portugal has been investing in hiking and cycling trail infrastructures to boost sustainable tourism. However, the lack of reliable data on the use of these trails means that the times of greatest affluence or the type of user who makes the most use of them are not recorded. These data are of the utmost importance to the managing bodies, with which they can adjust their actions to improve the management, maintenance, promotion, and use of the infrastructures for which they are responsible. The aim of this work is to present a review study on projects, techniques, and methods that can be used to identify and count the different types of users on these trails. The most promising computer vision techniques are identified and described: YOLOv3-Tiny, MobileNet-SSD V2, and FasterRCNN with ResNet-50. Their performance is evaluated and compared. The results observed can be very useful for proposing future prototypes. The challenges, future directions, and research opportunities are also discussed.
... These approaches demonstrate that lightweight networks can achieve excellent tracking performance. Similarly, Gomes et al. [16] proposed a novel method for computing the number of people and bicycles in an edge AI system using the Jetson Nano board. They utilized the YOLO network and V-IOU tracker to perform real-time counting of people and bicycles. ...
In recent years, human interference in seismic-station environments has posed challenges to the quality and accuracy of seismic signals, making data processing difficult. To accurately identify interference caused by personnel and ensure the reliability of seismic-network instrument detection data, it is necessary to track the detected targets across consecutive frames. Deep neural networks have made significant progress in this field. Therefore, an intelligent identification solution for environmental interference at seismic stations is proposed, which combines deep learning with multi-object tracking techniques. A centroid-matching tracking algorithm based on Kalman filtering is introduced to identify the entry/exit timestamps, alongside motion trajectories of interfering individuals, thereby marking the anomalous data caused by the presence of interfering personnel in seismic time-series data. Experimental results demonstrate that this research provides an effective solution for intelligent identification of environmental interference in seismic station environments.
... It is able to perform target detection and recognition, and predicts the positions and categories of multiple candidate objects at the same time [20]. It has been used in a wide range of applications for object detection, including real-time and near real-time applications [21,22]. For those reasons, this network was chosen for the present study, due to the high performance and ability to process large amounts of data in reasonable amounts of time. ...
Solar activity has been subject to increasingly more research in the last decades. Its influence on life on Earth is now better understood. Solar winds impact the earth’s magnetic field and atmosphere. They can disrupt satellite communication and navigation tools and even electrical power grids and several other infrastructure crucial for our technology-based society. Coronal mass ejections (CMEs), solar energetic particles, and flares are the main causes of problems that affect the systems mentioned. It is possible to predict some of those by monitoring the sun and analyzing the images obtained in different spectra, thus identifying solar phenomena related to its activity, such as filaments, pores, and sunspots. Several studies have already been carried out on the subject of automation of the mentioned analysis, most of which use neural networks and other machine learning approaches. In this work, we develop a method for sunspot detection based on the YOLOv5 network, applying it to a dataset of images from the Geophysical and Astronomical Observatory of the University of Coimbra (OGAUC), which has one of the oldest and more complete datasets of sun images in the world. Our method reaches mAP@.5 over 90% with YOLOv5s, which is higher than other methods previously applied for the same dataset. This shows that CNN models can be used in spectroheliographs for detecting and tracking sunspots.
... In addition, the system recognizes only the required objects, which increases the system's speed, unlike other systems that rely on detecting all elements simultaneously. In 2022 [11], the research presents an application based on the Yolo and Alfi algorithm for tracking, where the system counts human beings in addition to counting bicycles. The algorithm reduces the number of frames by selecting only the necessary frames, which increases the algorithm's power and reduces memory consumption. ...
Crowd detection has various applications nowadays. However, detecting humans in crowded circumstances is difficult because the features of different objects conflict, making cross-state detection impossible. Detectors in the overlapping zone may therefore overreact. In this paper, real-time people counting is proposed using a proposed model of the YOLOv5 (You Only Look Once) algorithm and KCF (kernel correlation filter) algorithm. The YOLOv5 algorithm was used because it is considered one of the most accurate algorithms for detecting people in real-time. Despite the high accuracy of the YOLOv5 algorithm in detecting the people in the image, video, or real-time camera capturing, it needs an increase in speed.
For this reason, the YOLOv5 algorithm was combined with the KCF tracking algorithm. Where the YOLOv5 algorithm identifies people to be tracked by the KCF. The YOLOv5 algorithm was trained on a database of people, and the system's accuracy reached 98%. The speed of the proposed system was increased after adding the KCF.
... In addition, the system recognizes only the required objects, which increases the system's speed, unlike other systems that rely on detecting all elements simultaneously. In 2022 [11], the research presents an application based on the Yolo and Alfi algorithm for tracking, where the system counts human beings in addition to counting bicycles. The algorithm reduces the number of frames by selecting only the necessary frames, which increases the algorithm's power and reduces memory consumption. ...
... • The lack of suitable models or training data sets for detection of application-specific concepts (i.e., objects, human poses) for the underground mining environment. • The difficulty of decision making on technology selection for realizing object detection ( [5], [6], [7], [8], [9], and [10]) and human pose recognition ( [11], [12], [13], and [14]) on edge computing devices with heterogeneous resource-constraints (e.g., different device variants of NVIDIA's Jetson devices). ...
... However, Yolov6 achieved higher processing performance in comparison to Yolov7 [7]. A selective frame-down sampling method has been developed for improving Yolo-based object detection (people counting application) performance with Jetson Nano-devices [10]. Particularly, Yolov5 models optimized with TensorRT achieved ~63-108 ms inference latency. ...
... The review indicates that many Yolo-based [5], [6], [7], and [8] object detection approaches have been developed. When targeting model execution on edge computing devices with limited resources, the models may be optimized/compressed (TF-Lite, TensorRT) [9] and [10] or partitioned [19], [20], [21], and [22] between edge and cloud environments. Many solutions [12], [27], and [28] have been developed for human pose recognition, also targeting execution on mobile devices [13] and [14]. ...
Realization of situation-awareness for autonomous robotics applications in edge computing environment is challenging. First, computing capabilities of edge devices are limited, which must be considered in the execution of machine learning (ML)-based solutions. Second, many technologies are available for realizing situation-aware capabilities, but comparison and integration of solutions creates additional challenges. Third, existing ML-based models are often not directly applicable for realizing custom applications, and model(s) may need to be re-trained with new data. The contribution of this paper is efficiency and feasibility evaluation of human pose recognition and object detection technologies in edge computing environment. Several lessons learnt covering constraints are presented regarding feasibility of the experimented technologies and data sets. The efficiency evaluation results indicated that simultaneous human pose recognition (Google’s Movenet) and object detection (Yolov5) on Jetson AGX Xavier achieved ~13-16 FPS, while GPU and CPU utilization remained at a medium level, and most of the memory remained unused (< 44 %). Object concept and human pose concept activation algorithms may be considered as an additional contribution. Realized architecture design of the prototyped system in multiple computing environments can be considered as a partial evaluation of a ML-based big data reference architecture.
