Aerial Video Research Articles

Aerial image recognition in discriminative bi-transformer

Aerial image recognition (AIR) has received increasing interest in the remote sensing community, and complex backgrounds with various types of objects bring huge challenges to recognition. Vision transformer (ViT) can capture long-range contextual relations, but it tends to fall into over-fitting for training samples. This problem is particularly acute in AIR where trainable data is insufficient. While training ViT with label smoothing can alleviate this problem, it is easy to pose the risk of underfitting and relies heavily on data quality. Furthermore, the application of ViT in AIR has always been limited because aerial images generally have greater intra-class variation and inter-class similarity than natural images. Thus, we propose a ViT-based framework called D-BiT to extend ViT to learn discriminative features in aerial images by integrating two heads on the ViT backbone. One is a classifier head with the Discriminative Label Smoothing Module (DLSM) to fully learn effective features without overfitting the distribution of noisy data, even in the condition of insufficient data volume. Another is a projector head with a newly-designed supervised contrastive loss to effectively combine label information and achieve a more compact and reasonable intra-class structure. Experimental results on three popular AIR benchmarks demonstrate the effectiveness of D-BiT: Aerial Image Dataset (AID), UC Merced and Event Recognition in Aerial videos (ERA), where we achieve state-of-the-art (SOTA) performance. In particular, D-BiT can significantly improve the accuracy of ViT for recognizing aerial images.

Research on Influencing Factors of Traffic Conflicts in Freeway Diverging Area during the Maintenance Period

Based on the traffic conflict techniques, a microscopic safety analysis framework is presented to study traffic conflicts in the freeway exit ramp diverging area (FERDA) during the maintenance period. A manual–software combination data processing approach was used to extract trajectory data from aerial videos and identify actual conflicts. Simulation conflicts were obtained from the VISSIM simulation model. Firstly, the conflict features of maintenance and normal operation periods were compared. Subsequently, conflicts were classified as lane-change conflicts or rear-end conflicts to investigate their spatiotemporal characteristics during the maintenance period. Finally, a random parameters multinomial logit model was developed to study the influencing factors and heterogeneity of conflicts. The results revealed that the FERDA has a higher collision risk during the maintenance period compared with the normal operation period. Additionally, lane-change conflicts have a higher occurrence probability than rear-end conflicts, but a shorter duration. Besides, the conflict density heat map showed that the lane-change conflicts and rear-end conflicts are densely distributed at 0–150 m and the middle lane of the FERDA, respectively. The model results indicated that the factors that significantly influence conflicts include velocity, acceleration, type, position, and destination of vehicle, the heavy vehicle percentage, total number of vehicles in the FERDA, and the velocity and destination difference between vehicles. Among these factors, the type of vehicle, destination difference, and the position of vehicles (at 150–270 m of the FERDA) were found that have significantly heterogeneous impacts on the two types of conflicts. This study is expected to provide safety analysts with an effective analysis method for evaluating the risk of the FERDA during the maintenance period. Additionally, the concluded influencing factors can potentially help develop different and effective safety countermeasures for different types of conflicts.

Aerial photogrammetry of seabirds from digital aerial video images using relative change in size to estimate flight height

Aerial photogrammetry of seabirds from digital aerial video images using relative change in size to estimate flight height

A Comparative Study on Deep Learning and Machine Learning Models for Human Action Recognition in Aerial Videos

Unmanned Aerial Vehicle )UAV( finds its significant application in video surveillance due to its low cost, high portability and fast-mobility. In this paper, the proposed approach focuses on recognizing the human activity in aerial video sequences through various keypoints detected on the human body via OpenPose. The detected keypoints are passed onto machine learning and deep learning classifiers for classifying the human actions. Experimental results demonstrate that multilayer perceptron and SVM outperformed all the other classifiers by reporting an accuracy of 87.80% and 87.77% respectively whereas LSTM did not produce very good results as compared to other classifiers. Stacked Long Short-Term Memory networks (LSTM( produced an accuracy of 71.30% and Bidirectional LSTM yielded an accuracy of 76.04%. The results also indicate that machine learning models performed better than deep learning models. The major reason for this finding is the lesser availability of data and the deep learning models being data hungry models require a large amount of data to work upon. The paper also analyses the failure cases of OpenPose by testing the system on aerial videos captured by a drone flying at a higher altitude. This work provides a baseline for validating machine learning classifiers and deep learning classifiers against recognition of human action from aerial videos.

Broad Federated Meta-Learning of Damaged Objects in Aerial Videos

Broad Federated Meta-Learning of Damaged Objects in Aerial Videos

Aerial-Based Weed Detection Using Low-Cost and Lightweight Deep Learning Models on an Edge Platform

Highlights Lightweight deep learning models were trained on an edge device to identify weeds in aerial images. A customized configuration file was setup to train the models. These models were deployed to detect weeds in aerial images and videos (near real-time). CSPMobileNet-v2 and YOLOv4-lite are recommended models for weed detection using edge platform. Abstract. Deep learning (DL) techniques have proven to be a successful approach in detecting weeds for site-specific weed management (SSWM). In the past, most of the research work has trained and deployed pre-trained DL models on high-end systems coupled with expensive graphical processing units (GPUs). However, only a limited number of research studies have used DL models on an edge system for aerial-based weed detection. Therefore, while focusing on hardware cost minimization, eight DL models were trained and deployed on an edge device to detect weeds in aerial-image context and videos in this study. Four large models, namely CSPDarkNet-53, DarkNet-53, DenseNet-201, and ResNet-50, along with four lightweight models, CSPMobileNet-v2, YOLOv4-lite, EfficientNet-B0, and DarkNet-Ref, were considered for training a customized DL architecture. Along with trained model performance scores (average precision score, mean average precision (mAP), intersection over union, precision, and recall), other model metrics to assess edge system performance such as billion floating-point operations/s (BFLOPS), frame rates/s (FPS), and GPU memory usage were also estimated. The lightweight CSPMobileNet-v2 and YOLOv4-lite models outperformed others in detecting weeds in aerial image context. These models were able to achieve a mAP score of 83.2% and 82.2%, delivering an FPS of 60.9 and 61.1 during near real-time weed detection in aerial videos, respectively. The popular ResNet-50 model achieved a mAP of 79.6%, which was the highest amongst all the large models deployed for weed detection tasks. Based on the results, the two lightweight models, namely, CSPMobileNet-v2 and YOLOv4-lite, are recommended, and they can be used on a low-cost edge system to detect weeds in aerial image context with significant accuracy. Keywords: Aerial image, Deep learning, Edge device, Precision agriculture, Weed detection.

Using Unoccupied Aerial Vehicles to estimate availability and group size error for aerial surveys of coastal dolphins

AbstractAerial surveys are frequently used to estimate the abundance of marine mammals, but their accuracy is dependent upon obtaining a measure of the availability of animals to visual detection. Existing methods for characterizing availability have limitations and do not necessarily reflect true availability. Here, we present a method of using small, vessel‐launched, multi‐rotor Unoccupied Aerial Vehicles (UAVs, or drones) to collect video of dolphins to characterize availability and investigate error surrounding group size estimates. We collected over 20 h of aerial video of dive‐surfacing behaviour across 32 encounters with Australian humpback dolphins Sousa sahulensis off north‐western Australia. Mean surfacing and dive periods were 7.85 sec (se = 0.26) and 39.27 sec (se = 1.31) respectively. Dolphin encounters were split into 56 focal follows of consistent group composition to which example approaches to estimating availability were applied. Non‐instantaneous availability estimates, assuming a 7 sec observation window, ranged between 0.22 and 0.88, with a mean availability of 0.46 (CV = 0.34). Availability tended to increase with increasing group size. We found a downward bias in group size estimation, with true group size typically one individual more than would have been estimated by a human observer during a standard aerial survey. The variability of availability estimates between focal follows highlights the importance of sampling across a variety of group sizes, compositions and environmental conditions. Through data re‐sampling exercises, we explored the influence of sample size on availability estimates and their precision, with results providing an indication of target sample sizes to minimize bias in future research. We show that UAVs can provide an effective and relatively inexpensive method of characterizing dolphin availability with several advantages over existing approaches. The example estimates obtained for humpback dolphins are within the range of values obtained for other shallow‐water, small cetaceans, and will directly inform a government‐run program of aerial surveys in the region.

Aircraft tracking in aerial videos based on fused RetinaNet and low‐score detection classification

Aircraft tracking in aerial videos is a challenge due to the small size and appearance affinity with other targets. Recently, the tracking-by-detection trackers, such as simple online and realtime tracking (SORT), attracted the attention of many researchers due to the great evolution in deep learning-based object detectors. SORT depends on motion cues in solving the data association problem by matching high-score detection bounding boxes to tracklets. However, motion cues are insufficient to track occluded or similar aircraft efficiently. In addition, most occluded aircrafts have a low-confidence percentage. The main objective of this paper is to propose an F-SORT tracker, an enhanced SORT tracker with the following improvements. Firstly, fused RetinaNet for small aerial targets is used as a detection framework to boost the detectability of SORT. Then, aircraft appearance features are extracted from the shallowest prediction layer and utilized as additional cues to improve the data association algorithm. Finally, the extracted features distinguish aircraft online from backgrounds via a K-nearest neighbour classifier. Experiments on an airport dataset show that F-SORT can significantly improve aircraft tracking. F-SORT outperforms other state-of-the-art trackers, achieving 72.75%, 59.63%, and 82.89% on MOTA, HOTA, and IDF1. In addition, F-SORT classifying algorithm boosts other trackers when applied.

Preliminary Evaluation of a System with On-Body and Aerial Sensors for Monitoring Working Dogs.

This paper presents a system for behavioral, environmental, and physiological monitoring of working dogs using on-body and aerial sensors. The proof of concept study presented here includes two trained dogs performing nine scent detection tasks in an uncontrolled environment encompassing approximately two acres. The dogs were outfitted with a custom designed wearable harness to monitor their heart rate, activity levels and skin temperature. We utilized a commercially available micro-air vehicle to perform aerial sensing by tracking the terrain and movement of the dog in the outdoor space. The dogs were free to explore the space working at maximal speeds to complete a scent-based search-and-retrieval task. Throughout the experiment, the harness data was transferred to a base station via Wi-Fi in real-time. In this work, we also focused on testing the performance of a custom 3D electrode with application specific ergonomic improvements and adaptive filter processing techniques to recover as much electrocardiography data as possible during high intensity motion activity. We were able to recover and use 84% of the collected data where we observed a trend of heart rate generally increasing immediately after successful target localization. For tracking the dogs in the aerial video footage, we applied a state-of-the-art deep learning algorithm designed for online object tracking. Both qualitative and quantitative tracking results are very promising. This study presents an initial effort towards deployment of on-body and aerial sensors to monitor the working dogs and their environments during scent detection and search and rescue tasks in order to ensure their welfare, enable novel dog-machine interfaces, and allow for higher success rate of remote and automated task performance.

Circles in the sea: annual courtship "torus" behaviour of basking sharks Cetorhinus maximus identified in the eastern North Atlantic Ocean.

Groups of basking sharks engaged in circling behaviour are rarely observed, and their function remains enigmatic in the absence of detailed observations. Here, underwater and aerial video recordings of multiple circling groups of basking sharks during late summer (August and September 2016-2021) in the eastern North Atlantic Ocean showed groups numbering between 6 and 23 non-feeding individuals of both sexes. Sharks swam slowly in a rotating "torus" (diameter range: 17-39 m), with individuals layered vertically from the surface to a maximum depth of 16 m. Within a torus, sharks engaged in close-following, echelon, close-flank approach or parallel-swimming behaviours. Measured shark total body lengths were 5.4-9.5m (mean LT : 7.3m ± 0.9s.d.; median: 7.2m, n= 27), overlapping known lengths of sexually mature males and females. Males possessed large claspers with abrasions that were also observed on female pectoral fins. Female body colouration was paler than that of males, similar to colour changes observed during courtship and mating in other shark species. Individuals associated with most other members rapidly (within minutes), indicating toroidal behaviours facilitate multiple interactions. Sharks interacted through fin-fin and fin-body contacts, rolling to expose the ventral surfaces to following sharks, and breaching behaviour. Toruses formed in late summer when feeding aggregations in zooplankton-rich thermal fronts switched to non-feeding following and circling behaviours. Collectively, the observations explain a courtship function for toruses. This study highlights northeast Atlantic coastal waters as a critical habitat supporting courtship reproductive behaviour of endangered basking sharks, the first such habitat identified for this species globally.

A Novel GAN-Based Anomaly Detection and Localization Method for Aerial Video Surveillance at Low Altitude

The last two decades have seen an incessant growth in the use of Unmanned Aerial Vehicles (UAVs) equipped with HD cameras for developing aerial vision-based systems to support civilian and military tasks, including land monitoring, change detection, and object classification. To perform most of these tasks, the artificial intelligence algorithms usually need to know, a priori, what to look for, identify. or recognize. Actually, in most operational scenarios, such as war zones or post-disaster situations, areas and objects of interest are not decidable a priori since their shape and visual features may have been altered by events or even intentionally disguised (e.g., improvised explosive devices (IEDs)). For these reasons, in recent years, more and more research groups are investigating the design of original anomaly detection methods, which, in short, are focused on detecting samples that differ from the others in terms of visual appearance and occurrences with respect to a given environment. In this paper, we present a novel two-branch Generative Adversarial Network (GAN)-based method for low-altitude RGB aerial video surveillance to detect and localize anomalies. We have chosen to focus on the low-altitude sequences as we are interested in complex operational scenarios where even a small object or device can represent a reason for danger or attention. The proposed model was tested on the UAV Mosaicking and Change Detection (UMCD) dataset, a one-of-a-kind collection of challenging videos whose sequences were acquired between 6 and 15 m above sea level on three types of ground (i.e., urban, dirt, and countryside). Results demonstrated the effectiveness of the model in terms of Area Under the Receiving Operating Curve (AUROC) and Structural Similarity Index (SSIM), achieving an average of 97.2% and 95.7%, respectively, thus suggesting that the system can be deployed in real-world applications.

Aerial orthoimage generation for UAV remote sensing: Review

With its unique advantages of high flexibility and high efficiency, UAV has become a reasonable substitute for conventional aerial measurement technology. Especially in the low altitude remote sensing image processing, the ortho-rectification and mosaic of aerial images are the key to vision-based UAV orthoimage generation. Therefore, how to select the appropriate methods to rectify and mosaic the aerial images of UAV is significance to research the automatic generation of digital orthoimages. Unfortunately, most of the existing reviews only focus on the general image mosaic techniques, and there are few special reports on the application of UAV orthoimage generation for reference. This paper presents a comprehensive survey on UAV orthoimage generation technologies. We conclude three mainstream frameworks of visual orthoimage generation, which are 2D mosaic framework based, SfM framework based and SLAM framework based methods. According to the above three specific frameworks, we first carried out a detailed description and comparative analysis of related important algorithms, and sorted out the differences, common points and inherent relationships. Considering the wide application of deep learning in UAV remote sensing, we propose some hypotheses on how to introduce deep learning technology into above three orthoimage generation frameworks. After analysis, we provide a more detailed performance quantification comparison of the two most recent potential frameworks (State-of-the-art methods based on SfM and SLAM). It is worth noting that we integrated different test data sources of UVA aerial video sequence with a general SLAM testing platform, and solve the issue that SLAM-based orthoimage generation methods are difficult to evaluate cross-platform. Finally, challenges about visual UAV orthoimage generation and future directions in addressing these challenges are also pointed out.

Predation in the Anthropocene: Harbour Seal (Phoca vitulina) Utilising Aquaculture Infrastructure as Refuge to Evade Foraging Killer Whales (Orcinus orca)

The rapid emergence of new marine developments (e.g., marine renewables, port infrastructure) alongside the substantial growth of existing industries has ultimately resulted in an unprecedented increase in anthropogenic structures within the marine environment over the previous century. Knowledge of whether marine species interact with, avoid, or accommodate and adapt to such structures is essential to ensure that further development of marine environments do not compromise conservation objectives of marine species. This article documents one such interaction. Herein, we describe the observation of a harbour seal (Phoca vitulina) seeking refuge from a group of foraging killer whales (Orcinus orca) within a blue (aka common) mussel (Mytilus edulis) farm. Aerial video footage (38 min 27 s) was collected using an unmanned aerial system during an encounter at an aquaculture site in Gulberwick Bay, Shetland, UK. Analysis of the footage showed the killer whale group spent 73.7% of the total encounter time exhibiting predatory associated behaviours and that they were observed interacting with the mussel farm infrastructure only during “predation activity” for a total of 26 min 52 s (72.8%). The harbour seal interacted with the mussel farm infrastructure during re- and proactive anti-predator behaviour and when exhibiting fatigue for 27 min 59 s, 94.4% of the total time the seal was observed. It is clear that both marine and terrestrial predator–prey interactions are increasingly occurring in settings that are in some way defined by the Anthropocene. The implications of this are discussed, including potential entanglement risk and human-altered “landscapes of fear.” As comprehension of the potential effects of human-altered risk grows, such knowledge should be taken into consideration prior to further modification of marine habitats.

SiamGauss: Siamese region proposal network with Gaussian head for visual object tracking

We propose SiamGauss, a Siamese region proposal network with a Gaussian head for single-target visual object tracking for aerial benchmarks. Visual tracking in aerial videos faces unique challenges due to the large field of view resulting in small size objects, similar looking objects (confusers) in close proximity, occlusions, and fast motion due to simultaneous object and camera motion. In Siamese tracking, a cross-correlation ration is performed in the embedding space to obtain a similarity map of the target within a search frame, which is then used to localize the target. The proposed Gaussian head helps suppress the activation produced in the similarity map on confusers present in the search frame during training while boosting the confidence on the target. This activation suppression improves the confuser awareness of our tracker. In addition, improving the activation on the target helps maintain tracking consistency in fast motion. Our proposed Gaussian head is only applied during training and introduces no additional computational overhead during inference while tracking. Thus, SiamGauss achieves fast runtime performance. We evaluate our method on multiple aerial benchmarks showing that SiamGauss performs favorably with state-of-the-art trackers while rating at a frame rate of 96 frames per second.

Vehicle Recognition from Unmanned Aerial Vehicle Videos Based on Fusion of Target Pre-Detection and Deep Learning

For accurate and effective automatic vehicle identification, morphological detection and deep convolutional networks were combined to propose a method for locating and identifying vehicle models from unmanned aerial vehicle (UAV) videos. First, the region of interest of the video frame image was sketched and grey-scale processing was performed; sub-pixel-level skeleton images were generated based on the Canny edge detection results of the region of interest; then, the image skeletons were decomposed and reconstructed. Second, a combination of morphological operations and connected domain morphological features were applied for vehicle target recognition, and a deep learning image benchmark library containing 244,520 UAV video vehicle samples was constructed. Third, we improved the AlexNet model by adding convolutional layers, pooling layers, and adjusting network parameters, which we named AlexNet*. Finally, a vehicle recognition method was established based on a candidate target extraction algorithm with AlexNet*. The validation analysis revealed that AlexNet* achieved a mean F1 of 85.51% for image classification, outperforming AlexNet (82.54%), LeNet (63.88%), CaffeNet (46.64%), VGG16 (16.67%), and GoogLeNet (14.38%). The mean values of Pcor, Pre, and Pmiss for cars and buses reached 94.63%, 6.87%, and 4.40%, respectively, proving that this method can effectively identify UAV video targets.

Unoccupied aerial video (UAV) surveys as alternatives to BRUV surveys for monitoring elasmobranch species in coastal waters

Abstract Effective conservation strategies are founded by baseline information on abundance and diversity estimates. Method choice can influence the success of baseline surveys as method performance is variable and needs to be selected based on habitat and taxa. Here, we assess the suitability of unoccupied aerial vehicle (UAV) surveys, specifically multi-rotor “drones”, and baited remote underwater video (BRUV) surveys in shallow-water habitats to quantify elasmobranch abundance and diversity in the Saudi Arabian central Red Sea. Our results show that the number of elasmobranchs h−1 observed using UAV surveys exceeded that of BRUV surveys by two orders of magnitude, indicating that the increased spatial coverage of UAV surveys is beneficial for long-term monitoring projects. BRUV surveys detected a greater number of species within reef habitats, whereas UAV surveys detected a greater number of species within sandflat habitats, indicating the value of multi-method approaches for regional biodiversity studies. Here, we provide the first insight into elasmobranchs associated with sandflat habitats in Saudi Arabia, emphasising the importance of these habitats to stingrays and the need for further information on elasmobranch habitat use to better inform management and conservation efforts in the face of rapid coastal developments across the Red Sea.

Multiscale anti-deformation network for target tracking in UAV aerial videos

With the popularity of unmanned aerial vehicle (UAV) technology, remote sensing target tracking in aerial videos from UAVs has drawn much attention. However, various problems occur, such as obvious target scale changes and frequent target shape updates in UAV aerial videos, and typical tracking algorithms have difficulties in solving these challenges. Therefore, we propose a remote sensing target tracking method for UAV videos based on a multiscale antideformation network (MSADN). This method uses the fully convolutional Siamese network (SiameseFC) as its basic architecture. First, in the target feature extraction stage, we use a receptive field block module to change the single convolutional layer of the network. And the multibranch structure can improve the algorithm’s adaptability to target scale changes. Then, in the tracking stage, to solve the problem of the lack of template updates, we integrate a template dynamic update module into the SiameseFC architecture. This module uses long short-term memory to generate control signals to dynamically update target shape information, and it will improve the algorithm’s ability to cope with frequent target shape updates. Finally, compared to state-of-the-art trackers, experimental results show that the MSADN algorithm can obtain better performance (75.9% Prec, 56.4% Succ) while ensuring higher efficiency (60 fps).

Developing a More Reliable Aerial Photography-Based Method for Acquiring Freeway Traffic Data

Due to the widespread use of unmanned aerial vehicles (UAVs) in remote sensing, there are fully developed techniques for extracting vehicle speed and trajectory data from aerial video, using either a traditional method based on optical features or a deep learning method; however, there are few papers that discuss how to solve the issue of video shaking, and existing vehicle data are rarely linked to lane lines. To address the deficiencies in current research, in this study, we formulated a more reliable method for real traffic data acquisition that outperforms the traditional methods in terms of data accuracy and integrity. First, this method implements the scale-invariant feature transform (SIFT) algorithm to detect, describe, and match local features acquired from high-altitude fixed-point aerial photographs. Second, it applies “you only look once” version 5 (YOLOv5) and deep simple online and real-time tracking (DeepSORT) to detect and track moving vehicles. Next, it leverages the developed Python program to acquire data on vehicle speed and distance (to the marked reference line). The results show that this method achieved over 95% accuracy in speed detection and less than 20 cm tolerance in vehicle trajectory mapping. This method also addresses common problems involving the lack of quality aerial photographic data and accuracy in lane line recognition. Finally, this approach can be used to establish a Frenet coordinate system, which can further decipher driving behaviors and road traffic safety.

Trajectory-based characteristic analysis and decision modeling of the lane-changing process in intertunnel weaving sections.

Existing lane-changing models generally neglect the detailed modeling of lane-changing actions and model lane-changing only as an instantaneous event. In this study, an intertunnel weaving section was taken as the background, the lane-changing duration and distance in the lane-changing process were taken as the main research objects. The detailed modeling of a lane-changing action was emphasized. Aerial videos of intertunnel weaving sections were collected, and accurate vehicle trajectory data were extracted. Basic data analysis shows that the lane-changing duration has a lognormal distribution and the lane-changing distance has a normal distribution. To analyze the difference of the lane-changing behavior characteristics in different lane-changing environments, based on the lead spacing and lag spacing in the target lane, a hierarchical clustering algorithm was applied to classify the lane-changing environment into six different types. Then, a deep neural network regression model was applied to model the lane-changing process for each environment type. The results show that the horizontal distribution, vertical distribution and statistical characteristics of the lane changing points under different lane-changing environments are significantly different. The prediction accuracy of the lane-changing distance after classification is improved by at least 61%, and the prediction accuracy of the lane-changing duration after classification is improved by at least 57%. It is also found that lane-changing behavior characteristics with large or small lag spacing are easier to predict, while in the other cases, the randomness of the lane-changing behavior characteristics is more obvious. The research results can be incorporated into lane-changing decision assistance systems and micro traffic simulation models to make the assistance system safer and more effective, and the simulation outputs should be more realistic and accurate.

Automated Parts-Based Model for Recognizing Human–Object Interactions from Aerial Imagery with Fully Convolutional Network

Advanced aerial images have led to the development of improved human–object interaction recognition (HOI) methods for usage in surveillance, security, and public monitoring systems. Despite the ever-increasing rate of research being conducted in the field of HOI, the existing challenges of occlusion, scale variation, fast motion, and illumination variation continue to attract more researchers. In particular, accurate identification of human body parts, the involved objects, and robust features is the key to effective HOI recognition systems. However, identifying different human body parts and extracting their features is a tedious and rather ineffective task. Based on the assumption that only a few body parts are usually involved in a particular interaction, this article proposes a novel parts-based model for recognizing complex human–object interactions in videos and images captured using ground and aerial cameras. Gamma correction and non-local means denoising techniques have been used for pre-processing the video frames and Felzenszwalb’s algorithm has been utilized for image segmentation. After segmentation, twelve human body parts have been detected and five of them have been shortlisted based on their involvement in the interactions. Four kinds of features have been extracted and concatenated into a large feature vector, which has been optimized using the t-distributed stochastic neighbor embedding (t-SNE) technique. Finally, the interactions have been classified using a fully convolutional network (FCN). The proposed system has been validated on the ground and aerial videos of the VIRAT Video, YouTube Aerial, and SYSU 3D HOI datasets, achieving average accuracies of 82.55%, 86.63%, and 91.68% on these datasets, respectively.