Lightweight Pedestrian Detection Network for UAV Remote Sensing Images Based on Strideless Pooling

While remarkable progress has been made to pedestrian detection in recent years, robust pedestrian detection in the wild e.g., under surveillance scenarios with occlusions, remains a challenging problem. In this paper, we present a novel approach for joint pedestrian and body part detection via semantic relationship learning under unconstrained scenarios. Specifically, we propose a Body Part Indexed Feature (BPIF) representation to encode the semantic relationship between individual body parts (i.e., head, head-shoulder, upper body, and whole body) and highlight per body part features, providing robustness against partial occlusions to the whole body. We also propose an Adaptive Joint Non-Maximum Suppression (AJ-NMS) to replace the original NMS algorithm widely used in object detection, leading to higher precision and recall for detecting overlapped pedestrians. Experimental results on the public-domain CUHK-SYSU Person Search Dataset show that the proposed approach outperforms the state-of-the-art methods for joint pedestrian and body part detection in the wild.

Pedestrian detection is another significant special application of object detection in autonomous vehicles. In contrast to universal object detection, it has similarities and special traits. Nevertheless, there are some difficulties that influence pedestrian detection performance, namely (i) occlusion and deformation (ii) low-quality and multispectral images consisting primarily of lighting conditions, small-scale detection, and target detection extensively, and (iii) true-false pedestrians. Deep learning (DL) methods are a class of artificial intelligence method that can solve the issues mentioned above of pedestrian detection. This paper initially gives an elaborate description of pedestrian detection, difficulties in pedestrian detection, and latest advancements in solving them using the assistance of DL methods with informative discussions, aiming to provide insights to the readers. (2) A new pedestrian detection algorithm (PDA) of true/false pedestrian is suggested here, in which a new YOLO-3D CNN model is applied to reject true/false pedestrian. The primary purpose is to evaluate the performance of the existing 3D CNN taking into consideration the problem of rejecting true false pedestrians based on images captured using the car's onboard cameras and light detection and ranging (LiDAR) sensors. PDA initially utilizes YOLOv3 to capture the entire image for training detector model capable of real-time forecasting. Next, as a feature extractor, it utilizes the MobileNet-SSD that provides great accuracy as well as good trading uptime. PDA then implements the Faster R-CNN method to detect different parts of the object, over the convolutional layer. Lastly, data augmentation techniques are applied in PDA to augment the data coverage by fully exploiting available training data. Simulation results indicate that the proposed pedestrian detection model and PDA improve the accuracy of real and false pedestrians while maintaining real-time requirements.

Due to the application requirement of single source tracking in dynamic background, it is difficult to get accurate pedestrian target detection. Through comparing performance quality of pedestrian describing operator between CENTRIST and HOG, we adopt CENTRIST feature extraction to combine with SVM off-line classifier and to train off-line model for target detection. Furthermore, we propose the importance to utilize edge classification thought to deepen contour feature. Meanwhile, we put forward the scheme to remove local texture and background noise to improve algorithm detection performance. The experiment offers detection performance comparison based on INRIA pedestrian detection data set in practical teachers' recording scene. It also explains that the improvement and optimization for CENTRIST pedestrian detection algorithm are very effective.

Vision-based pedestrian detection technology is widely used in security surveillance, such as transportation, prisons, and important places. Security monitoring requires pedestrian detection not only to be accurate and reliable, but also to meet the needs of real-time monitoring, and the false alarm rate is low. The existing pedestrian learning method based on deep learning has high reliability, but the timeliness cannot meet the monitoring requirements. Traditional pedestrian detection methods are time-sensitive, but false alarm rates tend to be higher. This paper proposes a self-learning pedestrian detection method, combined with motion detection, HOG feature extraction, and SVM classification for preliminary pedestrian detection. The false alarms caused during the pedestrian detection process are manually confirmed, and then the system automatically detects the false alarm image at a specific time. Self-learning, retraining the classifier, so as to continuously reduce the pedestrian detection false alarm rate and improve the pedestrian detection performance index through self-learning during the system operation, and the timeliness can meet the needs of video surveillance. Experiments show that the false alarm rate of this method is low, timeliness is high, and the recognition rate is also high.

In practice, the object detection algorithm is limited by a complex detection environment, hardware costs, computing power, and chip running memory. The performance of the detector will be greatly reduced during operation. Determining how to realize real-time, fast, and high-precision pedestrian recognition in a foggy traffic environment is a very challenging problem. To solve this problem, the dark channel de-fogging algorithm is added to the basis of the YOLOv7 algorithm, which effectively improves the de-fogging efficiency of the dark channel through the methods of down-sampling and up-sampling. In order to further improve the accuracy of the YOLOv7 object detection algorithm, the ECA module and a detection head are added to the network to improve object classification and regression. Moreover, an 864 × 864 network input size is used for model training to improve the accuracy of the object detection algorithm for pedestrian recognition. Then the combined pruning strategy was used to improve the optimized YOLOv7 detection model, and finally, the optimization algorithm YOLO-GW was obtained. Compared with YOLOv7 object detection, YOLO-GW increased Frames Per Second (FPS) by 63.08%, mean Average Precision (mAP) increased by 9.06%, parameters decreased by 97.66%, and volume decreased by 96.36%. Smaller training parameters and model space make it possible for the YOLO-GW target detection algorithm to be deployed on the chip. Through analysis and comparison of experimental data, it is concluded that YOLO-GW is more suitable for pedestrian detection in a fog environment than YOLOv7.

Summary The wellsite serves as the fundamental unit in the development of oil and gas fields, functioning as a hub for the production activities, with workover operations being a critical means to ensure production continuity. In addition, it plays a crucial role in environmental protection, preventing oil and gas leakage and pollution. Various pieces of mechanical equipment deployed at the wellsite are essential for tasks such as oil and gas extraction and well repair operations, holding a pivotal position in oil- and gasfield development. Consequently, intelligent wellsite implementation necessitates a primary focus on monitoring mechanical equipment, with video emerging as a vital form of multisource information at the wellsite. While existing research on wellsite video monitoring predominantly addresses system and data transmission issues, it falls short in addressing the challenges of real-time assessment and early warning in intelligent wellsite operations. This study introduces a method for identifying critical targets at the wellsite based on a scale-adaptive network. The model employs a multiscale fusion network to extract different image features and semantic features at various scales, facilitating their fusion. The processing of wellsite video images occurs in multiple stages, outputting predicted box locations and category information, enabling the localization and recognition of critical objects at the wellsite. Unlike traditional deep convolutional object detection methods, this model incorporates a parameter-free attention mechanism, enhancing the accurate feature learning of small targets during the extraction process and addressing the issue of multiscale imbalance. The experimental results validate the robust performance of the method, surpassing the latest one-stage object detection models and mainstream loss function methods. Comparative experiments demonstrate a 9.22% improvement in mean average precision (mAP) compared with YOLOv8, establishing the proposed model as a top performer in loss function optimization experiments. Furthermore, we propose a video security detection model whose results, combined with the recognition model, are applicable for video detection in wellsite scenarios. The model exhibits strong integration capabilities for scene area division and behavior regulation monitoring. In addition, the model provides valuable insights for analyzing equipment operating status, aligning with the practical needs of oil fields.

Pedestrian detection has a wide range of real-world critical applications including security and management of emergency scenarios. In critical applications, detection recall and precision are both essential to ensure the correct detection of all pedestrians. The development and deployment of object detection vision-based models is a time-consuming task, depending on long training and fine-tuning processes to achieve top performance. We propose an alternative approach, based on a fusion of pre-trained off-the-shelf state-of-the-art object detection models, and exploit base model divergences to quickly deploy robust ensembles with improved performance. Our approach promotes model reuse and does not require additional learning algorithms, making it suitable for rapid deployments of critical systems. Experimental results, conducted on PASCAL VOC07 test dataset, reveal mean average precision (mAP) improvements over base detection models, regardless of the set of models selected. Improvements in mAP were observed starting from just two detection models and reached 3.53% for a fusion of four detection models, resulting in an absolute fusion mAP of 83.65%. Moreover, the hyperparameters of our ensemble model may be adjusted to set an appropriate tradeoff between precision and recall to fit different recall and precision application requirements.

Computer vision techniques have been frequently applied to pedestrian and cyclist detection for the purpose of providing sensing capabilities to autonomous vehicles, and delivery robots among other use cases. Most current computer vision approaches for pedestrian and cyclist detection utilize RGB data alone. However, RGB-only systems struggle in poor lighting and weather conditions, such as at night, or during fog or precipitation, often present in pedestrian detection contexts. Thermal imaging presents a solution to these challenges as its quality is independent of time of day and lighting conditions. The use of thermal imaging input, such as those in the Long Wave Infrared (LWIR) range, is thus beneficial in computer vision models as it allows the detection of pedestrians and cyclists in variable illumination conditions that would pose challenges for RGB-only detection systems. In this paper, we present a pedestrian and cyclist detection method via thermal imaging using a deep neural network architecture. We have evaluated our proposed method by applying it to the KAIST Pedestrian Benchmark dataset, a multispectral dataset with paired RGB and thermal images of pedestrians and cyclists. The results suggest that our method achieved an F1-score of 81.34%, indicating that our proposed approach can successfully detect pedestrians and cyclists from thermal images alone.

Illumination, for pedestrian detection at nighttime is weak, and detection is easily affected through variations in illumination. Thus, a bicharacteristic method of pedestrian detection at nighttime based on hierarchical tree cascade classification is presented according to “coarse-to-fine” principle. The proposed method consists of two stages of cascade classifiers. Coarse cascade classifiers are constructed in complete binary tree architecture. These classifiers use Haar-like features for the rapid identification of candidate pedestrian areas. By contrast, fine cascade classifiers have a parallel structure. Edgelet features are used for detection along three parts: the head-shoulder, trunk, and leg parts of candidate pedestrian areas. Bayesian decision-making is adopted to achieve pedestrian target detection and a comprehensive analysis of the detection results from these three parts. Experiments show that the proposed method has high accuracy, ideal real-time performance, and strong reliability...

In the rapid development of deep learning, YOLO, as the first popular single-stage object detection model, has sparked an innovative storm in the computer vision community with its remarkable architecture and innovative concepts, marking a significant leap forward in object detection technology. Today, it is not only regarded as a milestone in this field but also sets an unparalleled example in the pursuit of the perfect fusion of detection speed and accuracy. YOLO has been widely applied in various fields such as agriculture, industry, pedestrian detection, and more. This research project will first introduce traditional object detection methods, then analyze object detection based on deep learning, and subsequently elaborate on the fundamental concepts of YOLO. It will systematically sort through the YOLO family and its significant improvements. Finally, based on different improvement strategies or application scenarios, the YOLO algorithm will be systematically classified and summarized.

Object detection has gradually developed into a popular research area because of the rife use of Remote Sensing Images (RSIs) in military and civil domains. However, complex backgrounds and problems with small items are the two biggest obstacles in object detection. Deep learning techniques have a significant advantage for object detection over conventional methods that rely on manually derived characteristics, and they have a lot of attention. Due to the wide range of objects and complex visual backgrounds in RSIs, current deep‐learning algorithms in the area of RSI object detection leave much to be desired. For this case, algorithms need to be specifically optimized. In this paper, a novel optimization‐driven Enhanced Faster Region Convolutional Neural Network (FRCNN) is proposed for object detection. This approach has five modules, namely pre‐processing, data augmentation, segmentation, feature extraction, and object detection. The detection process is performed using Enhanced FRCNN and it is trained by the proposed Gazelle White Shark Optimization Algorithm (GWSOA). Extensive experiments in high‐resolution RSI data sets have exposed the efficacy of the proposed approach. The novel approach achieved better accuracy of 97.31%, Mean Average Precision (MAP) of 97.85%, precision of 97.76%, recall of 96.16%, F ‐score of 95.78%, and error rate of 2.69.

The continuing development of remote sensing has resulted in a rapidly increasing number of remote sensing applications. High-resolution remote sensing images are used in various fields in the military. We propose methods for object detection based on remote sensing images. We develop a signal processing method for normalizing remote sensing images to eliminate noise such as fog, haze, and poor lighting. This method further improves detection accuracy and reduces error rates. We develop YOLOv4-faster, an accelerated neural network model based on the YOLO (You-Only-Look-Once) object detection method. YOLOv4-faster outperforms existing networks in terms of execution time and detection performance. We conduct a series of experiments on two public datasets (TGRS-HRRSD and NWPU VHR-10) as well as a dataset containing six military target classes provided by IMINT & Analysis and collected from Google Earth. YOLOv4-faster improves efficiency by utilizing multi-scale operations for the accurate detection of objects of various sizes, especially small objects. The experimental results show improved mAP (mean average precision) performance of the proposed method for object detection in remote sensing images. We thus propose a novel system for automatic object detection for high-resolution remote sensing images.

Applications of remote sensing images in both defense and civilian sectors have spurred substantial research interest. In the field of remote sensing, object detection confronts challenges such as complex backgrounds, scale diversity, and the presence of dense small objects. To address these issues, we propose an improved deep learning-based model, the Global Multi-scale Fusion Self-calibration Network, which is expected to contribute to alleviating the challenges. It consists of three main components: the hierarchical feature aggregation backbone, which uses improved modules such as the receptive field context-aware feature extraction module, the global information acquisition module, and the simple parameter-free attention module to extract key features and minimize the background interference. To couple multi-scale features, we enhanced the fusing component and designed the multi-scale enhanced pyramid structure integrating the proposed new modules. During the detection phase, especially when focusing on small object detection, we designed a novel convolutional attention feature fusion head. This head is constructed to integrate local and global branches for feature extraction by leveraging channel shuffling and multi-head attention mechanisms for efficient and accurate detection. Experiments on the Detection in Optical Remote Sensing Images (DIOR), Northwestern Polytechnical University Very High-Resolution‐10 (NWPU VHR‐10), remote sensing object detection (RSOD), and DOTAv1.0 data sets show that our method achieves mAP50(mean average precision at 50% intersection over union) of 69.7%, 91.3%, 94.2%, and 70.0%, respectively, outperforming existing comparative methods. The proposed network is expected to provide new perspectives for remote sensing tasks and possible solutions for relevant applications in the image domain.

To solve the low accuracy and poor stability in traditional object tracking methods for martial arts competition videos, a Kalman filtering algorithm based on feature matching and multi object tracking is proposed for object detection in martial arts competition videos. Firstly, feature matching in multi target tracking is studied. Then, based on target feature matching, the Kalman filtering algorithm is fused to construct a target detection model in martial arts videos. Finally, simulation experiments are conducted to verify the performance and application effectiveness of the model. The results showed that the average tracking errors of the model on the X and Y axes were 3.86% and 3.38%, respectively. At the same time, the average accuracy and recall rate in the video target tracking process were 93.64% and 95.48%, respectively. After 100 iterations, the results gradually stabilized. This indicated that the constructed model could accurately detect targets in martial arts competition videos. It had high tracking accuracy and robustness. Compared with traditional object detection methods, this algorithm has better performance and effectiveness. The Kalman filter algorithm based on feature matching and multi target tracking has broad application prospects and research value in target detection in martial arts competition videos.

The monitoring of grape quality parameters within viticulture using airborne remote sensing is an increasingly important aspect of precision viticulture. Airborne remote sensing allows high volumes of spatial consistent data to be collected with improved efficiency over ground-based surveys. Spectral data can be used to understand the characteristics of vineyards, including the characteristics and health of the vines. Within viticultural remote sensing, the use of cover-crop spectra for monitoring is often overlooked due to the perceived noise it generates within imagery. However, within viticulture, the cover crop is a widely used and important management tool. This study uses multispectral data acquired by a high-resolution uncrewed aerial vehicle (UAV) and Sentinel-2 MSI to explore the benefit that cover-crop pixels could have for grape yield and quality monitoring. This study was undertaken across three growing seasons in the southeast of England, at a large commercial wine producer. The site was split into a number of vineyards, with sub-blocks for different vine varieties and rootstocks. Pre-harvest multispectral UAV imagery was collected across three vineyard parcels. UAV imagery was radiometrically corrected and stitched to create orthomosaics (red, green, and near-infrared) for each vineyard and survey date. Orthomosaics were segmented into pure cover-cropuav and pure vineuav pixels, removing the impact that mixed pixels could have upon analysis, with three vegetation indices (VIs) constructed from the segmented imagery. Sentinel-2 Level 2a bottom of atmosphere scenes were also acquired as close to UAV surveys as possible. In parallel, the yield and quality surveys were undertaken one to two weeks prior to harvest. Laboratory refractometry was performed to determine the grape total acid, total soluble solids, alpha amino acids, and berry weight. Extreme gradient boosting (XGBoost v2.1.1) was used to determine the ability of remote sensing data to predict the grape yield and quality parameters. Results suggested that pure cover-cropuav was a successful predictor of grape yield and quality parameters (range of R2 = 0.37–0.45), with model evaluation results comparable to pure vineuav and Sentinel-2 models. The analysis also showed that, whilst the structural similarity between the both UAV and Sentinel-2 data was high, the cover crop is the most influential spectral component within the Sentinel-2 data. This research presents novel evidence for the ability of cover-cropuav to predict grape yield and quality. Moreover, this finding then provides a mechanism which explains the success of the Sentinel-2 modelling of grape yield and quality. For growers and wine producers, creating grape yield and quality prediction models through moderate-resolution satellite imagery would be a significant innovation. Proving more cost-effective than UAV monitoring for large vineyards, such methodologies could also act to bring substantial cost savings to vineyard management.