Adaptive Computation Offloading for Mobile Augmented Reality

Augmented reality (AR) underpins many emerging mobile applications, but it increasingly requires more computation power for better machine understanding and user experience. While computation offloading promises a solution for high-quality and interactive mobile AR, existing methods work best for high-definition videos but cannot meet the real-time requirement for emerging 4K videos due to the long uploading latency. We introduce ACTOR, a novel computation-offloading framework for 4K mobile AR. To reduce the uploading latency, ACTOR dynamically and judiciously downscales the mobile video feed to be sent to the remote server. On the server-side, it leverages image super-resolution technology to scale back the received video so that high-quality object detection, tracking and rendering can be performed on the full 4K resolution. ACTOR employs machine learning to predict which of the downscaling resolutions and super-resolution configurations should be used, by taking into account the video content, server processing delay, and user expected latency. We evaluate ACTOR by applying it to over 2,000 4K video clips across two typical WiFi network settings. Extensive experimental results show that ACTOR consistently and significantly outperforms competitive methods for simultaneously meeting the latency and user-perceived video quality requirements.

An Integrated Approach for Monitoring Social Distancing and Face Mask Detection Using Stacked ResNet-50 and YOLOv5

SARS-CoV-19 is one of the deadliest pandemics the world has witnessed, taking around 5,049,374 lives till now across worldwide and 459,873 in India. To limit its spread numerous countries have issued many safety measures. Though vaccines are available now, still face mask detection and maintain social distance are the key aspects to prevent from this pandemic. Therefore, authors have proposed a real-time surveillance system that would take the input video feed and check whether the people detected in the video are wearing a mask, this research further monitors the humans for social distancing norms. The proposed methodology involves taking input from a CCTV feed and detecting humans in the frame, using YOLOv5. These detected faces are then processed using Stacked ResNet-50 for classification whether the person is wearing a mask or not, meanwhile, DBSCAN has been used to detect proximities within the persons detected.

Social Distance Analysing - COVID-19 Using Deep Learning and Computer Vision

In the combat in opposition to the coronavirus, social distancing has tested to be an effective degree to bog down the unfold of the disease. The machine provided is for reading social distancing through calculating the space among humans for you to gradual down the unfold of the virus. This machine makes use of enter from video frames to parent out the space among people to relieve the impact of this pandemic. This is performed through comparing a video feed acquired through a surveillance camera. The video is calibrated into bird’s view and fed as an enter to the YOLOv3 version that is an already educated item detection version. The YOLOv3 version is educated using the Common Object in Context (COCO). The proposed machine turned into corroborated on a pre-filmed video. The outcomes and consequences acquired through the machine display that assessment of the space among more than one people and figuring out if policies are violated or not. If the space is less than the minimal threshold value, the people are represented through a purple bounding box, if not then it's far represented through a inexperienced bounding box. This machine may be similarly advanced to detect social distancing in real-time applications.

Re-Identification in Urban Scenarios: A Review of Tools and Methods

With the widespread use of surveillance image cameras and enhanced awareness of public security, objects, and persons Re-Identification (ReID), the task of recognizing objects in non-overlapping camera networks has attracted particular attention in computer vision and pattern recognition communities. Given an image or video of an object-of-interest (query), object identification aims to identify the object from images or video feed taken from different cameras. After many years of great effort, object ReID remains a notably challenging task. The main reason is that an object’s appearance may dramatically change across camera views due to significant variations in illumination, poses or viewpoints, or even cluttered backgrounds. With the advent of Deep Neural Networks (DNN), there have been many proposals for different network architectures achieving high-performance levels. With the aim of identifying the most promising methods for ReID for future robust implementations, a review study is presented, mainly focusing on the person and multi-object ReID and auxiliary methods for image enhancement. Such methods are crucial for robust object ReID, while highlighting limitations of the identified methods. This is a very active field, evidenced by the dates of the publications found. However, most works use data from very different datasets and genres, which presents an obstacle to wide generalized DNN model training and usage. Although the model’s performance has achieved satisfactory results on particular datasets, a particular trend was observed in the use of 3D Convolutional Neural Networks (CNN), attention mechanisms to capture object-relevant features, and generative adversarial training to overcome data limitations. However, there is still room for improvement, namely in using images from urban scenarios among anonymized images to comply with public privacy legislation. The main challenges that remain in the ReID field, and prospects for future research directions towards ReID in dense urban scenarios, are also discussed.

Situation Awareness in Remote Operators of Autonomous Vehicles: Developing a Taxonomy of Situation Awareness in Video-Relays of Driving Scenes

Even entirely driverless vehicles will sometimes require remote human intervention. Existing SA frameworks do not acknowledge the significant human factors challenges unique to a driver in charge of a vehicle that they are not physically occupying. Remote operators will have to build up a mental model of the remote environment facilitated by monitor view and video feed. We took a novel approach to “freeze and probe” techniques to measure SA, employing a qualitative verbal elicitation task to uncover what people “see” in a remote scene when they are not constrained by rigid questioning. Participants (n = 10) watched eight videos of driving scenes randomized and counterbalanced across four road types (motorway, rural, residential and A road). Participants recorded spoken descriptions when each video stopped, detailing what was happening (SA Comprehension) and what could happen next (SA Prediction). Participant transcripts provided a rich catalog of verbal data reflecting clear interactions between different SA levels. This suggests that acquiring SA in remote scenes is a flexible and fluctuating process of combining comprehension and prediction globally rather than serially, in contrast to what has sometimes been implied by previous SA methodologies (Jones and Endsley, 1996; Endsley, 2000, 2017b). Inductive thematic analysis was used to categorize participants’ responses into a taxonomy aimed at capturing the key elements of people’s reported SA for videos of driving situations. We suggest that existing theories of SA need to be more sensitively applied to remote driving contexts such as remote operators of autonomous vehicles.

Iterative Approach on Redesigning and Construction of a Pipe Inspection Robot

<p>A pipe inspection robot is a device that is inserted into pipes to check for obstructions or damage. These robots are traditionally manufactured offshore, are extremely expensive, and are often not adequately supported in the event of malfunction. This had resulted in Associated Environmental Services Limited, a New Zealand utiliser of this equipment, facing significant periods of down-time as they wait for their robots to be repaired. Recently, they were informed that several of their robots were no longer supported. At their instigation, this project was conceived to redesign the electronics and control system of one of these pipe inspection robots, utilising the existing mechanical platform. Requirements for the robot were that it must operate reliably in confined, dark and wet environments, and provide a human user with a digital video feed of the internal status of the pipes. This robot should, as much as possible, incorporate off-the-shelf components, facilitating cheap, and potentially on-site repair. This project details the redesign and construction of such a robot. It employs three electronic boards integrated with the mechanical components and provides video feedback via a custom graphical user interface. Although at the prototype stage, the electronic redesign has been successful, with a cost of less than a tenth of the original robot purchase price.</p>

Iterative Approach on Redesigning and Construction of a Pipe Inspection Robot

<p>A pipe inspection robot is a device that is inserted into pipes to check for obstructions or damage. These robots are traditionally manufactured offshore, are extremely expensive, and are often not adequately supported in the event of malfunction. This had resulted in Associated Environmental Services Limited, a New Zealand utiliser of this equipment, facing significant periods of down-time as they wait for their robots to be repaired. Recently, they were informed that several of their robots were no longer supported. At their instigation, this project was conceived to redesign the electronics and control system of one of these pipe inspection robots, utilising the existing mechanical platform. Requirements for the robot were that it must operate reliably in confined, dark and wet environments, and provide a human user with a digital video feed of the internal status of the pipes. This robot should, as much as possible, incorporate off-the-shelf components, facilitating cheap, and potentially on-site repair. This project details the redesign and construction of such a robot. It employs three electronic boards integrated with the mechanical components and provides video feedback via a custom graphical user interface. Although at the prototype stage, the electronic redesign has been successful, with a cost of less than a tenth of the original robot purchase price.</p>

COMPUTER VISION BASED TRAFFIC RULE BREACH DETECTION

This paper is about the detection of traffic rule breach via computer vision which takes the feed from the traffic surveillance system, processes the video feed, detects the breach and alerts the traffic police. The number of traffic accidents is on the rise with the increasing number of vehicles. Traffic breach is the biggest cause of accidents. So, to mitigate this problem our system processes the CCTV camera feed in real-time, detects the traffic rule breach events and sends the push notification to the android based application of the traffic police stationed nearby; so, further actions can be taken. As this system detects breach faster than humans, the concerned authoritarian department will be at ease in implementing safe roads accurately. This system acts as an add-on to the current video surveillance system rather than building new infrastructure. Thus, the output of this system can be used not only or safety and security purposes but as well as for analytical purposes with effective traffic monitoring at a lower cost. Hence, this system aids law enforcement agencies in implementing road safety efficiently and effectively ensuring smooth traffic flow.

Face Mask Detection in Real-Time using MobileNetv2

Face mask detection has made considerable progress in the field of computer vision since the start of the Covid-19 epidemic. Many efforts are being made to develop software that can detect whether or not someone is wearing a mask. Many methods and strategies have been used to construct face detection models. A created model for detecting face masks is described in this paper, which uses “deep learning”, “TensorFlow”, “Keras”, and “OpenCV”. The MobilenetV2 architecture is used as a foundation for the classifier to perform real-time mask identification. The present model dedicates 80 percent of the training dataset to training and 20% to testing, and splits the training dataset into 80% training and 20% validation, resulting in a final model with 65 percent of the dataset for training, 15 percent for validation, and 20% for testing. The optimization approach used in this experiment is “stochastic gradient descent” with momentum (“SGD”), with a learning rate of 0.001 and momentum of 0.85. The training and validation accuracy rose until they reached their maximal peak at epoch 12, with 99% training accuracy and 98% validation accuracy. The model's training and validation losses both reduced until they reached their lowest at epoch 12, with a validation loss of 0.050% and a training loss of less than 0.025%. This system allows for real-time detection of someone is missing the appropriate face mask. This model is particularly resource-efficient when it comes to deployment, thus it can be employed for safety. So, this technique can be merged with embedded application systems at public places and public services places as airports, trains stations, workplaces, and schools to ensure subordination to the guidelines for public safety. The current version is compatible with both IP and non-IP cameras. Web and desktop apps can use the live video feed for detection. The program can also be linked to the entrance gates, allowing only those who are wearing masks to enter. It can also be used in shopping malls and universities.

Direct and Indirect vSLAM Fusion for Augmented Reality

Augmented reality (AR) is an emerging technology that is applied in many fields. One of the limitations that still prevents AR to be even more widely used relates to the accessibility of devices. Indeed, the devices currently used are usually high end, expensive glasses or mobile devices. vSLAM (visual simultaneous localization and mapping) algorithms circumvent this problem by requiring relatively cheap cameras for AR. vSLAM algorithms can be classified as direct or indirect methods based on the type of data used. Each class of algorithms works optimally on a type of scene (e.g., textured or untextured) but unfortunately with little overlap. In this work, a method is proposed to fuse a direct and an indirect methods in order to have a higher robustness and to offer the possibility for AR to move seamlessly between different types of scenes. Our method is tested on three datasets against state-of-the-art direct (LSD-SLAM), semi-direct (LCSD) and indirect (ORBSLAM2) algorithms in two different scenarios: a trajectory planning and an AR scenario where a virtual object is displayed on top of the video feed; furthermore, a similar method (LCSD SLAM) is also compared to our proposal. Results show that our fusion algorithm is generally as efficient as the best algorithm both in terms of trajectory (mean errors with respect to ground truth trajectory measurements) as well as in terms of quality of the augmentation (robustness and stability). In short, we can propose a fusion algorithm that, in our tests, takes the best of both the direct and indirect methods.