- Deep Learning-Based Real-Time Multiple-Object Detection and Tracking from Aerial Imagery via a Flying Robot with GPU-Based Embedded Devices
Keywords
Multi-target detection and tracking; multi-copter drone; aerial imagery; image sensor; GPU-based embedded module; neural computing stick; image processing
In deep learning method real time object detection and tracking two embedded modules were used one was designed using Jetson TX or AGX Xavier and the other based on Intel neural compute stick. Then GPU- based embedded computing, effective target tracking and a tracking algorithm Deep SORT integrating DL based association metrics. Finally, we demonstrate a small multi-rotor drone validate the effectiveness of the proposed approach.
Implementation plan
Step 1: Initially, create the embedded module using a Jetson TX or AGX Xavier for real-time onboard computing power on small flying drones.
Step 2: Next, use the GPU-based embedded computing modules to obtain detailed metric data about frame rates, as well as the computation power.
Step 3: Next we perform tracking moving objects based on the extension of simple online and real-time tracking by using Deep SORT with YOLOv3.
Step 4: Next we track the target position using a GPU-based algorithm based on YOLOv2.
Step 5: The performance of these work is measured through the following performance metrics, power consumption and error rate.
Software Requirement: Python – 3.11.3 and Operating System: Windows 10(64-bit)
- Object Detection and Image Segmentation with Deep Learning on Earth Observation Data: A Review-Part I: Evolution and Recent Trends
Keywords
Artificial intelligence AI; machine learning; convolutional neural networks CNN; image segmentation; object detection; Earth observation
Deep learning has new challenges in the field of Earth observation (EO) the entry barriers for EO researchers are high due to the advances in Computer vision (CV) to lower this they focus on Image segmentation and object detection using CNN. Then we evaluate the performance of DL architecture on datasets as well as advances made in CV and their impact on future research.
Implementation plan
Step 1: Initially we load the dataset with bounding box labels for object detection on Microsoft Common Objects in Context (MS-COCO) test-dev set.
Step 2: Next, extracts feature from input data in a hierarchical manner based on convolutional backbone.
Step 3: Next, perform the image analysis based on two-stage detectors by using Region based CNN (R-CNN).
Step 4: Next, perform the process of one-stage detectors by using YOLO-V1 Detect the object centres and predicts object class and bounding box.
Step 5: The performance of these work is measured through the following performance metrics, Accuracy, Precision, Recall, ROC curves and F-Score.
Software Requirement: Python – 3.11.3 and Operating System: Windows 10(64-bit)
- Object Detection Binary Classifiers methodology based on deep learning to identify small objects handled similarly: Application in video surveillance
Keywords
Detection, One-Versus-All, One-Versus-One
The small objects manipulated with hand are essential in many fields, except video surveillance. In this paper the detection of small objects handled using binarization techniques. They introduce two level methodology based on DL called Object Detection with Binary Classifiers. The first level identifies candidate region in input frame and the second level employs CNN based binarization One-Versus-All or One-Versus-One.
Implementation plan
Step 1: Initially we load the input video with six objects: pistol, knife, smartphone, bill, purse and card.
Step 2: Next, selects the candidate regions from the input frame based on ODeBiC methodology.
Step 3: Next we apply binarization technique based on a CNN classifier with One-Versus-All or One-Versus-One based on ODeBiC methodology.
Step 4: Next, based on the classification detect the objects.
Step 5: The performance of these work is measured through the following performance metrics, False positives rate, Accuracy, Precision, Recall, ROC curves and F-Score.
Software Requirement: Python – 3.11.4 and Operating System: Windows 10(64-bit)
- Salient Object Detection in the Deep Learning Era: An In-depth Survey
Keywords
Computer Vision and Pattern Recognition
Salient object detection (SOD) is an essential problem in computer vision; it focuses on recent advances in SOD by DL based solutions (named Deep SOD) and covers various aspects including network architecture, level of supervision, learning paradigm, and object-level detection. SOD datasets with rich attributes covers various salient object types and the robustness to random input perturbations and adversarial attacks.
Implementation plan
Step 1: Initially we load the HRSOD SOD dataset.
Step 2: Next, perform the pixel-based attribute analysis.
Step 3: Next, perform the segmentation process for select the region of interest.
Step 4: Next, detect the object from the ROI parameters.
Step 5: The performance of this work is measured through the following performance metrics, MAE, Fbw, Recall, ROC curves and F- measure.
Software Requirement: Python – 3.11.4 and Operating System: Windows 10(64-bit)
- Deep learning-based object detection in low-altitude UAV datasets: A survey
Keywords
Unmanned aerial vehicles, Computer vision, Low-altitude aerial datasets
Object detection implemented in DL framework processes moving images captured from drones. UAV have interest in potential applications such as surveillance, visual navigation, object detection, and sensors-based obstacle. It can analyse recent contributions of these algorithms to low altitude UAV datasets. In this paper object detection algorithms can be divided two stages one is YOLO, SSD, RetinaNet and other is performance on low altitude UAV datasets.
Implementation plan
Step 1: Initially we load the low-altitude UAV datasets.
Step 2: Next, generate a sparse set of regions of interest (RoIs) for make a classification.
Step 3: Next, perform the process of localizing, classifying and predicting bounding boxes using a unified shared deep network.
Step 4: Next, perform the object detection process by using the deep learning algorithm fast RCNN
Step 5: The performance of this work is measured through the following performance metrics, Accuracy, mean area precision (mAP), Recall, ROC curves and F- measure.
Software Requirement: Python – 3.11.4 and Operating System: Windows 10(64-bit)
- Deep learning-based anomalous object detection system for panoramic cameras managed by a Jetson TX2 board
Keywords
Video surveillance, panoramic cameras, power saving, Solvents, Streaming media, Cameras, Hardware, Generators
In this paper DL-based foreground anomalous object detection system in video streams supplied by panoramic cameras, specially designed for video surveillance systems. This system optimizes new potential detection generator by three different multivariant homoscedastic distributions, it can reduce the need for GPU based hardware. Jetson TX2 board generate the better performance.
Implementation plan
Step 1: Initially we load the video streams, which is supplied by panoramic cameras.
Step 2: Next, design the Jetson TX2 board for make anomalous object detection.
Step 3: Next, build power efficient video surveillance systems in the hardware device.
Step 4: Next, perform the process of searching for anomalous objects through a new potential detection generator managed by three different multivariate homoscedastic distributions.
Step 5: The performance of this work is measured through the following performance metrics, Accuracy, Precision, Recall and F-Score.
Software Requirement: Python – 3.9.3 and Operating System: Windows 10(64-bit)
- A Survey of Modern Deep Learning based Object Detection Models
Keywords
Object recognition, lightweight networks
DL based object detection is the task of classification and localization of object in video or image. It has gained prominence due to its widespread applications. It uses benchmark datasets and evaluation metrics and also, they provide prominent backbone architectures for edge devices. It also covers contemporary lightweight classification models and at last we compare the performances of these architectures on multiple metrics.
Implementation plan
Step 1: Initially we load ImageNet Large Scale Visual Recognition datasets.
Step 2: in the two stage object detectors proposed the Feature Pyramid Network (FPN), for extract the image features and construct the pyramid and detect the objects in the image
Step 3: in the Single stage object detectors extract the image features and detect the small objects by using Single Shot MultiBox Detector (SSD).
Step 4: The performance of these work is measured through the following performance metrics, frames per second (FPS), precision and recall. However and mean Average Precision (mAP).
Software Requirement: Python – 3.11.3 and Operating System: Windows 10(64-bit)
- SaliencyGAN: Deep Learning Semisupervised Salient Object Detection in the Fog of IoT
Keywords
Internet of Things, Salient Object Detection, GAN.
This paper proposes a semisupervised adversarial learning method SaliencyGAN, designed for SOD in IOT. SaliencyGAN is empowered by a novel concatenated-GAN and the CNN can be chosen flexibly based on devices and applications. It uses both labelled and unlabelled data from different problem domains for training. SaliencyGAN were more robust to the common “mode missing” issue compared GAN models.
Implementation plan
Step 1: Initially, construct the SaliencyGAN Architecture with fog and IoT environment then we load the Salient object detection (SOD) dataset
Step 2: Next generate the multiple Convolutional layer-based process by using VGG16 structure
Step 3: Next we perform the training process based on the image generator GI , the image discriminator DI , the saliency generator GS and the saliency discriminator DS by using minibatch gradient descent.
Step 4: Next we perform the object detection based on the trained layer structure.
Step 5: The performance of these work is measured through the following performance metrics, pixel-wise mean absolute error (MAE) of saliency maps, maximum F-measure scores (maxF), Precision Recall curves (PR-curves), F-measure curves.
Software Requirement: Python – 3.11.3 and Operating System: Windows 10(64-bit)
- Object Detection with Deep Learning: A Review
Keywords
DL based object detection uses its representative tool namely CNN. Then we focus on typical generic object detection architecture to improve detection performance. They also cover several specific tasks, including salient object detection, face detection and pedestrian detection. At last several promising directions and tasks are provided for guidelines.
Implementation plan
Step 1: Initially we load the input images from the dataset PASCAL VOC 2012.
Step 2: Next predicts bounding boxes directly from locations of the topmost feature map by using R-CNN.
Step 3: Next we perform the CNN based deep feature extraction process to extract a 4096- dimensional feature.
Step 4: Next we perform the Classification by pre-trained category specific linear SVMs for multiple classes.
Step 5: The performance of these work is measured through the following performance metrics, Accuracy, Precision, Recall, ROC curves and F-Score.
Software Requirement: Python – 3.9.6 and Operating System: Windows 10(64-bit)
In case if you run short of time get your research writing service at phdservices.org, you get the quickest and the most accurate professional support. Our elegant research writing team and our diligent programmers work effectively and finish your deep learning project successfully.