YOLO based Human Action Recognition and Localization

Shinde, Shubham; Kothari, Ashwin; Gupta, Varun

doi:10.1016/j.procs.2018.07.112

Cited by 147 publications

(62 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…2 CNN: Image features are extracted by convolution operation, according to color, edge, texture and so on [16].…”

Section: Baseline Methodsmentioning

confidence: 99%

“…It uses the sum of the squared errors (SSE) as the loss function [15]. The YOLO algorithm divides images into S*S grids, and the output of each grid is (B*5+C) dimensions, which include the location information, the confidence of the border box [16], and the number of categories. However, these factors have different effects on the accuracy of the object recognition of vehicles.…”

Section: Loss Functionmentioning

confidence: 99%

See 1 more Smart Citation

A Novel Object Detection Algorithm in Video

2019

Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering

View full text Add to dashboard Cite

Deep learning technology performs effect in feature extraction of images. Nowadays, with the development of video monitoring, the application of deep learning technology to surveillance video has profound implications. The effects of traditional video recognition are not satisfactory, but deep learning methods perform effect in many scenes of image classification. This paper proposed a novel object detection algorithm in video. It combined the traditional methods of extracting feature and deep learning algorithm to realize vehicle identification based on surveillance video. The method used the frame difference method and background subtraction to preprocess the image, and then trained a network model based on YOLO to perform object detection and obtain the categories and location information of the monitored vehicle. Compared with the existing object detection algorithms faster RNN, our method can achieve higher accuracy and can significantly shorten the time for detection, which can recognize the object of vehicle video quickly and efficiently. The method can meet the requirements of real-time detection.

show abstract

“…2 CNN: Image features are extracted by convolution operation, according to color, edge, texture and so on [16].…”

Section: Baseline Methodsmentioning

confidence: 99%

Section: Loss Functionmentioning

confidence: 99%

A Novel Object Detection Algorithm in Video

2019

Proceedings of 2019 the 9th International Workshop on Computer Science and Engineering

View full text Add to dashboard Cite

show abstract

“…Several types of neural networks exist and the YOLO is the one chosen for this project, for the following reasons [17,18]:…”

Section: Neural Networkmentioning

confidence: 99%

“…Neural Network training A first training has been carried out using a tiny YOLOv2 pretrained with the COCO dataset. YOLO requires some files to start training which are [17,18] a configuration file with all layers of YOLO architecture (described in Figure 2). • pre-trained convolutional weights.…”

Section: 42mentioning

confidence: 99%

Augmented Reality and Deep Learning towards the Management of Secondary Building Assets

Corneli¹,

Naticchia²,

Carbonari³

et al. 2019

Proceedings of the International Symposium on Automation and Robotics in Construction (IAARC)

View full text Add to dashboard Cite

The retrieval of as-is information for existing buildings is a prerequisite for effectively operating facilities, through the creation or updating of Building/Asset Information Models (BIM/AIM), or Digital Twins. At present, many studies focus on the capture of geometry for the modelling of primary components, overlooking the fact that many recurring actions need to be conducted on assets inside buildings. Furthermore, highly accurate survey techniques like laser scanning need long offsite processing for object recognition. Performing such process on site would dramatically impact efficiency and also prevent the need to revisit the site in the case of insufficient/incomplete data. In this paper, an Augmented Reality (AR) system is proposed enabling inventory, information retrieval and information update directly on-site. It would reduce post-processing work and avoid loss of information and unreliability of data. The system has a Head-Mounted Display (HMD) AR interface that lets the technician interact handsfree with the real world and digital information contained in the BIM/AIM. A trained Deep Learning Neural Network operates the automatic recognition of objects in the field of view of the user and their placement into the digital BIM. In this paper, two uses cases are described: one is the inventory of small assets inside buildings to populate a BIM/AIM, and the second is the retrieval of relevant information from the AIM to support maintenance operations. Partial development and feasibility tests of the first use case applied to fire extinguishers, have been carried out to assess the feasibility and value of this system.

show abstract

“…It uses a separate CNN model to realize end-to-end target detection, divides the input images into 7 × 7 grids, and then each cell is responsible for predicting the targets in which the center points fall in the grid; when the pumping unit or head working fall in some grid, this grid is responsible for predicting them, compares the predicted value with the real value, and calculates the predicted loss. The core idea is to directly manipulate the whole picture by inputting a figure directly in the output layer for each grid to predict the B bounding box location information and the confidence score of the bounding box [26].…”

Section: Using Yolov3 As a Detector Of The Ylts Framework To Detect Tmentioning

confidence: 99%

Design and Analysis of Refined Inspection of Field Conditions of Oilfield Pumping Wells Based on Rotorcraft UAV Technology

Zhou

Wu³

et al. 2019

Electronics

View full text Add to dashboard Cite

The traditional oil well monitoring method relies on manual acquisition and various high-precision sensors. Using the indicator diagram to judge the working condition of the well is not only difficult to establish but also consumes huge manpower and financial resources. This paper proposes the use of computer vision in the detection of working conditions in oil extraction. Combined with the advantages of an unmanned aerial vehicle (UAV), UAV aerial photography images are used to realize real-time detection of on-site working conditions by real-time tracking of the working status of the head working and other related parts of the pumping unit. Considering the real-time performance of working condition detection, this paper proposes a framework that combines You only look once version 3 (YOLOv3) and a sort algorithm to complete multi-target tracking in the form of tracking by detection. The quality of the target detection in the framework is the key factor affecting the tracking effect. The experimental results show that a good detector makes the tracking speed achieve the real-time effect and provides help for the real-time detection of the working condition, which has a strong practical application.

show abstract

YOLO based Human Action Recognition and Localization

Cited by 147 publications

References 4 publications

A Novel Object Detection Algorithm in Video

A Novel Object Detection Algorithm in Video

Augmented Reality and Deep Learning towards the Management of Secondary Building Assets

Design and Analysis of Refined Inspection of Field Conditions of Oilfield Pumping Wells Based on Rotorcraft UAV Technology

Contact Info

Product

Resources

About