Camera Placement Meeting Restrictions of Computer Vision

Aghajanzadeh, Sara; Naidu, R. Ch. A.; Chen, Shuo-Han; Tung, Caleb; Goel, Abhinav; Lu, Yung-Hsiang; Thiruvathukal, George K.

doi:10.1109/icip40778.2020.9190851

Cited by 16 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given an image of an object, the goal is to return a group of images containing the queried object [1]. Systems employing object reID can be used to enhance public safety [2], manage crowds [3,4], and detect events across multiple cameras [5]. In these applications, object reID systems may be deployed on embedded devices like traffic cameras and drones [6].…”

Section: Introductionmentioning

confidence: 99%

Low-power object counting with hierarchical neural networks

Goel

Tung

Aghajanzadeh

et al. 2020

Proceedings of the ACM/IEEE International Symposium on Low Power Electronics and Design

Self Cite

View full text Add to dashboard Cite

Low-power computer vision on embedded devices has many applications. This paper describes a low-power technique for the object re-identification (reID) problem: matching a query image against a gallery of previously-seen images. State-ofthe-art techniques rely on large, computationally-intensive Deep Neural Networks (DNNs). We propose a novel hierarchical DNN architecture that uses attribute labels in the training dataset to perform efficient object reID. At each node in the hierarchy, a small DNN identifies a different attribute of the query image. The small DNN at each leaf node is specialized to re-identify a subset of the gallery-only the images with the attributes identified along the path from the root to a leaf. Thus, a query image is reidentified accurately after processing with a few small DNNs. We compare our method with state-of-the-art object reID techniques. With a ∼4% loss in accuracy, our approach realizes significant resource savings: 74% less memory, 72% fewer operations, and 67% lower query latency, yielding 65% less energy consumption.

show abstract

Section: Introductionmentioning

confidence: 99%

Low-power object counting with hierarchical neural networks

Goel

Tung

Aghajanzadeh

et al. 2020

Proceedings of the ACM/IEEE International Symposium on Low Power Electronics and Design

Self Cite

View full text Add to dashboard Cite

show abstract

“…The task reports the number of occurrences of the queried category of objects in the image [3,4]. Object counting can be used in automatic traffic control [5], crowd management [6], and wildlife monitoring [4]. Most existing techniques require large DNNs that are compute and memory intensive to achieve high accuracy [7].…”

Section: Introductionmentioning

confidence: 99%

Low-Power Object Counting with Hierarchical Neural Networks

Goel,

Tung,

Aghajanzadeh

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Deep Neural Networks (DNNs) can achieve state-ofthe-art accuracy in many computer vision tasks, such as object counting. Object counting takes two inputs: an image and an object query and reports the number of occurrences of the queried object. To achieve high accuracy on such tasks, DNNs require billions of operations, making them difficult to deploy on resource-constrained, low-power devices. Prior work shows that a significant number of DNN operations are redundant and can be eliminated without affecting the accuracy. To reduce these redundancies, we propose a hierarchical DNN architecture for object counting. This architecture uses a Region Proposal Network (RPN) to propose regions-of-interest (RoIs) that may contain the queried objects. A hierarchical classifier then efficiently finds the RoIs that actually contain the queried objects. The hierarchy contains groups of visually similar object categories. Small DNNs are used at each node of the hierarchy to classify between these groups. The RoIs are incrementally processed by the hierarchical classifier. If the object in an RoI is in the same group as the queried object, then the next DNN in the hierarchy processes the RoI further; otherwise, the RoI is discarded. By using a few small DNNs to process each image, this method reduces the memory requirement, inference time, energy consumption, and number of operations with negligible accuracy loss when compared with the existing object counters.

show abstract

“…Adjustable Camera Model. Some work [15,[105][106][107][108][109][110][111][112][113][114] study deployment problem with adjustable camera model. Some of them consider the direction only adjustable camera model, e.g., [105,106], which assume that sensors have been deployed in advance and design various scheduling mechanisms on cameras' directions.…”

Section: Camera Model -The Specific Directional Model For Uavmentioning

confidence: 99%

“…Some of them consider the direction only adjustable camera model, e.g., [105,106], which assume that sensors have been deployed in advance and design various scheduling mechanisms on cameras' directions. Some of them consider both focal length and direction adjustable camera [15,[107][108][109][110][111][112][113][114]. Specifically, [15] assumes the cameras are pre-deployed and focuses on adjusting the direction and focal length to maximize the number of covered objects.…”

Section: Camera Model -The Specific Directional Model For Uavmentioning

confidence: 99%

“…[108] utilizes the adjustable camera model to maximize the covered portion of a given area. [107] and [109] both build monitoring model based on the pixel requirement of computer vision application, e.g., face detection, and design the photo selection and coverage algorithm based on their model, respectively. [114] and [110] both study the deployment in 3D space.…”

Section: Camera Model -The Specific Directional Model For Uavmentioning

confidence: 99%

See 1 more Smart Citation

MoniSys: Sensing, Communication, and Video Analysis in UAV Monitoring System

Wang

View full text Add to dashboard Cite

Over the past 20 years, visual data has taken over every aspect of our lives, and camera deployment has experienced an unprecedented increase. For example, in the USA and UK, there is one camera for every 8 people used for diverse applications, like surveillance and public safety. However, in many extreme conditions, pre-deploying cameras are not feasible. Fortunately, the development of UAVs let these agile, flexible, and powerful devices make up the limitation of pre-deployed cameras. While high-resolution visual data offers rich information about the sensing environment, it causes significant challenges to the data analysis. Advances in computer vision present an excellent opportunity to process and analyze this massive amount of data; however, they have come at the expense of compute and network costs. For analysis, computing-resource-limited UAVs need to transmit sensing data to a computingresource-rich server (on the ground). This distributed architecture posits several network-level resource management challenges: to ensure optimal UAV trajectories for sensing visual data; and to address the mobility impact and fair data delivery in multi-UAV access networks; and to provide low-latency, high-accuracy, and lowbandwidth-cost analysis.We begin by presenting a general algorithm design schema for the waypoint planning problem to generate waypoints (i.e., UAVs hovering and sensing points) achieving quality bounded sensing data. This schema includes three steps: discretization divides the entire solution space into subspaces; then dominating set extraction find out all the optimal solutions in every subspace; at last, transform waypoint planning into submodular optimization problem and propose an approximate algorithm. We apply this schema to three scenarios and verify the performance of our method provides 1.6× gain in sensing data quality.Next, in order to address communication challenges, specifically the mobility impact and fairness among multiple UAV-server streamings, we develop VSiM -an easy-deployment and high-compatibility end-to-end solution to fairness in multiple mobile video streaming applications with a shared bottleneck bandwidth. It is pluggable to the server directly without caring and modifying any existing protocols or components. VSiM consists of three key techniques: dynamic and fair bandwidth allocation by incorporating mobile profile and QoE-related information; quick buffer filling for clients with lower playback time according to the requirement of the buffersensitive clients; adaptiveness to heterogeneous wireless network environments, like varied mobility patterns and topologies of base stations (BSes). It improves more iv than 40% on min QoE, which equals resolution improvement of viewing quality from 720p to 1080p) compared to state-of-the-art solutions.Finally, in order to achieve low-latency, high-accuracy, and low-bandwidth-cost analysis, we present AccDecoder -a new decoder that derives important video content from bits stream and enhances them by super-resolution (SR) model. SR m...

show abstract

Camera Placement Meeting Restrictions of Computer Vision

Cited by 16 publications

References 8 publications

Low-power object counting with hierarchical neural networks

Low-power object counting with hierarchical neural networks

Low-Power Object Counting with Hierarchical Neural Networks

MoniSys: Sensing, Communication, and Video Analysis in UAV Monitoring System

Contact Info

Product

Resources

About