Human Detection Task Research Articles

Disaster zone human and animal detection using sonar

The research introduces a novel approach to animal and human detection using sonar technology in fields such as disaster management, and under water exploration. Unlike traditional visual methods, sonar systems emit soundwaves to analyze echoes, providing unique advantages in challenging environments. The proposed method involves collecting raw sonar data, followed by preprocessing techniques for noise reduction, signal normalization, and feature extraction. Sonar’s ability to penetrate various media, including water and dense fog, makes it valuable for detecting animals and humans in low-visibility conditions. Additionally, sonar operates effectively in both day and night settings, unaffected by lighting conditions. The proposed detection system will undergo comprehensive experiments using representative datasets and real-world scenarios. Performance metrics, such as detection accuracy, precision, recall, and computational efficiency, will be analyzed and compared with existing approaches. The study showcases the effectiveness and viability of employing sonar technology for animal and human detection tasks, highlighting its unique capabilities in challenging environments.

Human detection on antistatic floors

Nowadays, a correct detection of people is very important for different purposes. Most applications use images as sources of information. However, an image may contain more information than is necessary for the detection task. For this reason, raw video images can end up being used for malicious purposes or privacy concerns. We present baseline results for a new human detection task. We evaluate long short-term memory (LSTM)-based deep learning models for detecting people using electrical signals from electrostatic discharge (ESD) floors as a source of information. Statistical features were provided to the models every second and four classification problems were studied. The first model discriminates between motion and non-motion. A second model classifies the action of the person between: no person, walking or standing. A third model classifies between walking and standing. And a last model predicts whether there is someone or no one on the ESD floor. Mattews Correlation Coefficient (MCC) was used as the main metric to evaluate the performance of the models. The LSTM models obtained a MCC between 0.94 and 0.99.

A hybrid LBP-DCNN based feature extraction method in YOLO: An application for masked face and social distance detection.

COVID-19 is an ongoing pandemic and the WHO recommends at least one-meter social distance, and the use of medical face masks to slow the disease’s transmission. This paper proposes an automated approach for detecting social distance and face masks. Thus, it aims to help the reduction of diseases transferred by respiratory droplets such as COVID-19. For this system, a two-cascaded YOLO is used. The first cascade detects humans in the environment and computes the social distance between them. Then, the second cascade detects human faces with or without a mask. Finally, red bounding boxes encircle the people’s images that did not follow the rules. Also, in this paper, we propose a two-part feature extraction approach used with YOLO. The first part of the proposed feature extraction method extracts general features using the transfer learning approach. The second part extracts better features specific to the current task using the LBP layer and classification layers. The best average precision for the human detection task was obtained as 66% using Resnet50 in YOLO. The best average precision for the mask detection was obtained as 95% using Darknet19+LBP with YOLO. Also, another popular object detection network, Faster R-CNN, have been used for comparison purpose. The proposed system performed better than the literature in human and mask detection tasks.

Human Detection Algorithm Based on Improved YOLO v4

The human behavior datasets have the characteristics of complex background, diverse poses, partial occlusion, and diverse sizes. Firstly, this paper adopts YOLO v3 and YOLO v4 algorithms to detect human objects in videos, and qualitatively analyzes and compares detection performance of two algorithms on UTI, UCF101, HMDB51 and CASIA datasets. Then, this paper proposed an improved YOLO v4 algorithm since the vanilla YOLO v4 has incomplete human detection in specific video frames. Specifically, the improved YOLO v4 introduces the Ghost module in the CBM module to further reduce the number of parameters. Lateral connection is added in the CSP module to improve the feature representation capability of the network. Furthermore, we also substitute MaxPool with SoftPool in the primary SPP module, which not only avoids the feature loss, but also provides a regularization effect for the network, thus improving the generalization ability of the network. Finally, this paper qualitatively compares the detection effects of the improved YOLO v4 and primary YOLO v4 algorithm on specific datasets. The experimental results show that the improved YOLO v4 can solve the problem of complex targets in human detection tasks effectively, and further improve the detection speed.

Vision-based Human Detection by Fine-Tuned SSD Models

Human-robot interaction (HRI) and human-robot collaboration (HRC) has become more popular as the industries are taking initiative to idealize the era of automation and digitalization. Introduction of robots are often considered as a risk due to the fact that robots do not own the intelligent as human does. However, the literature that uses deep learning technologies as the base to improve HRI safety are limited, not to mention transfer learning approach. Hence, this study intended to empirically examine the efficacy of transfer learning approach in human detection task by fine-tuning the SSD models. A custom image dataset is developed by using the surveillance system in TT Vision Holdings Berhad and annotated accordingly. Thereafter, the dataset is partitioned into the train, validation, and test set by a ratio of 70:20:10. The learning behaviour of the models was monitored throughout the fine-tuning process via total loss graph. The result reveals that the SSD fine-tuned model with MobileNetV1 achieved 87.20% test AP, which is 6.1% higher than the SSD fine-tuned model with MobileNetV2. As a trade-off, the SSD fine-tuned model with MobileNetV1 attained 46.2 ms inference time on RTX 3070, which is 9.6 ms slower as compared to SSD fine-tuned model with MobileNetV2. Taking test AP as the key metric, SSD fine-tuned model with MobileNetV1 is considered as the best fine-tuned model in this study. In conclusion, it has shown that the transfer learning approach within the deep learning domain can help to protect human from the risk by detecting human at the first place.

Human Detection via Image Denoising for 5G‐Enabled Intelligent Applications

5G technology strongly supports the development of various intelligent applications, such as intelligent video surveillance and autonomous driving. And the human detection technology in intelligent video surveillance has also ushered in new challenges. A number of video images will be compressed for efficient transmission; the resulting incomplete feature representation of images will drop the human detection performance. Therefore, in this work, we propose a new human detection method based on compressed denoising. We exploit the quality factor in the compressed image and incorporate the pixel_shuffle inverse transform based on FFDNet to effectively improve the performance of image compression denoising, then HRNet and HRFPN are used to extract and fuse high‐resolution features of denoised images, respectively, to obtain high‐quality feature representation, and finally, a cascaded object detector is used for classification and bounding box regression to further improve object detection performance. At last, the experimental results on PASCAL VOC show that the proposed method effectively removes the compression noise and further detects human objects with multiple scales and different postures. Compared with the state‐of‐the‐art methods, our method achieved better detection performance and is, therefore, more suited for human detection tasks.

Optimizing Sector Ring Histogram of Oriented Gradients for human injured detection from drone images

Developing a system for emergency response and rescue to find injured people within images has been an interest to many researchers. The key challenge is to define a proper feature to describe the human body's appearance. Various features are often extracted from low-level data, such as texture and colour (Zhang et al., Sensors. 19(5):1005, 2020). One of the strong features is the Sector Ring Histogram of Oriented Gradients (SRHOG) that has been successfully applied to human detection tasks. Despite good accuracy in finding humans, an SRHOG detection method produces a large amount of false-negative labels. Locating an injured body after a disaster in drone images using an imaging camera remained a challenge. This study presents a new extension to SRHOG, so-called AdSRHOG, to reduce the number of false labels. In our approach, the gradient filters used by SRHOG defined adaptively depending on the corresponding pixel location The proposed feature was used the Support Vector Machine (SVM) algorithm to detect humans on drone images. The experiments showed a significant improvement of up to 54.3% in reducing the false labels. It was also found that the overall accuracy of the human detection process had a notable improvement of 13.1% over a traditional SRHOG detection technique.

Human detection and tracking with deep convolutional neural networks under the constrained of noise and occluded scenes

Human detection and tracking is a key aspect in surveillance system due to its importance in timely identification of person, recognition of human activity and scene analysis. Convolutional neural networks have been widely used approach in detection and tracking related tasks. In this paper, a robust framework is presented for the human detection and tracking in noisy and occluded environments with the aid of data augmentation techniques. In addition, a softmax layer and integrated loss function is used to improve the detection and classification performance of the proposed model. The primary focus is to perform the human detection task in unconstrained environments. The implemented system outperforms the state-of-the-arts methods which can be validated from the experimental results.

Training Physical and Geometrical Mid-Points for Multi-person Pose Estimation and Human Detection Under Congestion and Low Resolution.

This paper introduces the design and evaluation of NeoPose which is developed for multi-person pose estimation and human detection. The design of NeoPose is targeting the issue of human detection under congested situation and with low resolution in the image. Under such situations, we compared the performance of different versions of NeoPose as well as other existing algorithms in a human detection task. Throughout the task, the usefulness of two kinds of mid-point (physical and geometrical mid-points) and a deconvolution structure was discussed. Experiment results indicated that NeoPose which applied geometrical mid-points and deconvolution structure performed the best in terms of both precision and recall in the evaluation.

Real-Time Human Detection for Aerial Captured Video Sequences via Deep Models.

Human detection in videos plays an important role in various real life applications. Most of traditional approaches depend on utilizing handcrafted features which are problem-dependent and optimal for specific tasks. Moreover, they are highly susceptible to dynamical events such as illumination changes, camera jitter, and variations in object sizes. On the other hand, the proposed feature learning approaches are cheaper and easier because highly abstract and discriminative features can be produced automatically without the need of expert knowledge. In this paper, we utilize automatic feature learning methods which combine optical flow and three different deep models (i.e., supervised convolutional neural network (S-CNN), pretrained CNN feature extractor, and hierarchical extreme learning machine) for human detection in videos captured using a nonstatic camera on an aerial platform with varying altitudes. The models are trained and tested on the publicly available and highly challenging UCF-ARG aerial dataset. The comparison between these models in terms of training, testing accuracy, and learning speed is analyzed. The performance evaluation considers five human actions (digging, waving, throwing, walking, and running). Experimental results demonstrated that the proposed methods are successful for human detection task. Pretrained CNN produces an average accuracy of 98.09%. S-CNN produces an average accuracy of 95.6% with soft-max and 91.7% with Support Vector Machines (SVM). H-ELM has an average accuracy of 95.9%. Using a normal Central Processing Unit (CPU), H-ELM's training time takes 445 seconds. Learning in S-CNN takes 770 seconds with a high performance Graphical Processing Unit (GPU).

Human detection in complex real scenes based on combination of biorthogonal wavelet transform and Zernike moments

Human detection in real scenes with high complexity is a crucial problem in computer vision research. For tackling the human detection task, several existing methods adopt one feature or combination of features to detect human objects. In this work, we propose a new combination of features based algorithm for human detection, which identifies the presence of a human, in complex real scenes. In the combination, the two features (i) biorthogonal wavelet transform (BWT) and (ii) Zernike moments (ZM) have been used. The approximate shift-invariance and symmetry properties of BWT facilitate the human detection in the wavelet domain. Specifically, the shift-invariance property of BWT is effective for translated object representation whereas the symmetry property yields perfect reconstruction for retaining object boundaries (i.e., edges). Moreover, translation and rotation-invariance properties of ZM are especially beneficial for the representation of varying pose and orientation of the human objects. For these reasons, the composite of the two features brings about significant synthesized benefits over each single feature and the other widely used features. In the experiments for human detection, we used two classifiers, AdaBoost and support vector machines, respectively, for the comparative study purpose, and the standard INRIA dataset and DaimlerChrysler dataset were used for the evaluations. Experimental results demonstrated the significant outperformance of the proposed method through quantitative evaluations and also suggest that the proposed hybridization of features is preferable for the classification problem.

Up-Fusion

The amount of multimedia data on the Internet has increased exponentially in the past few decades and this trend is likely to continue. Multimedia content inherently has multiple information sources, therefore effective fusion methods are critical for data analysis and understanding. So far, most of the existing fusion methods are static with respect to time, making it difficult for them to handle the evolving multimedia content. To address this issue, in recent years, several evolving fusion methods were proposed, however, their requirements are difficult to meet, making them useful only in limited applications. In this article, we propose a novel evolving fusion method based on the online portfolio selection theory. The proposed method takes into account the correlation among different information sources and evolves the fusion model when new multimedia data is added. It performs effectively on both crisp and soft decisions without requiring additional context information. Extensive experiments on concept detection and human detection tasks over the TRECVID dataset and surveillance data have been conducted and significantly better performance has been obtained.

Effective Feature Evaluation Method Using Enhanced SVM for Human Detection

SVM이론은 기계학습이론 분야에서 가장 강력한 학습 이론으로 사용되어 왔고 일반적으로 안정적인 성능을 가진 것으로 알려져 있다. 특정 분류 작업을 수행할 경우 SVM이론의 성공여부는 그 클래스를 대표하는 특징에 따라 상당한 영향을 받게 된다. 특정 클래스를 대표하는 특징이나 차별적인 특징이 주어졌을 때 SVM 학습과정은 최상의 일반론을 제시할 수 있고, 그 결과로 최상의 분류기가 될 수 있다. 본 논문에서는 휴먼 검출 분야에서 효과적인 특징 선택 문제를 평가하고 선택된 특징들의 성능을 평가하기 위한 효과적인 알고리즘을 제안한다. 본 논문에서는 특징 그룹을 생성하기 위해 새로운 군집 멤버로서 다중 크기를 가진 HOG-Family 특징을 제안한다. 또한 특징들의 차원을 줄이고 처리속도를 개선하기 위해 PCA이론과 SVM이론을 결합한 새로운 PCA-SVM이론을 제안한다. 제안한 방법은 대부분의 고유 특징 벡터는 유지하면서 처리 속도 및 특징점의 차원을 현저히 줄여주는 장점을 가지고 있다. 실험결과를 통해 제안한 알고리즘의 우수성을 입증해 보인다.

A Comparison of Planar Versus Volumetric Numerical Observers for Detection Task Performance in Whole-Body PET Imaging

The goal of this project is to determine if there is a difference between planar and volumetric numerical observers. The motivation comes from the use of a volumetric (3-D) display of image data (i.e., three orthogonal views through a volume image) to assess whole-body PET images in most, if not all, clinical PET centers. Studies of correlations between numerical observers and human observers (e.g., using receiver operating characteristic (ROC) detection analysis), however, typically use a planar (2-D) display and analysis methodology. In particular, planar implementations of the nonprewhitening matched filter (NPWMF) and the channelized Hotelling observer (CHO) have been extensively investigated as surrogates for measuring human detection task performance. To determine if there is a difference between planar and volumetric numerical observers, we analyzed the behavior of 2-D and 3-D implementations of the NPWMF and CHO numerical observers with multiple realizations of images with noise properties similar to those of whole-body PET oncology imaging. The results indicate that there is a significant increase in SNR or detectability of volumetric numerical observers relative to planar observers. This implies that volumetric and planar numerical observers may have different strengths of correlation with human observer performance, and that these correlations should be determined prior to using numerical observers to optimize algorithm or protocol performance for whole-body PET imaging. When axial smoothing is applied, however, to impose isotropic target resolution or to simulate fully 3-D PET data, the detectabilities of planar numerical observers are increased and the differences between planar and volumetric numerical observers become less significant.