Deep Learning for Risk Detection and Trajectory Tracking at Construction Sites

Zhao, Yu; Chen, Quan; Cao, Wengang; Ye, Jie; Xiong, Jian; Gui, Guan

doi:10.1109/access.2019.2902658

Cited by 61 publications

(29 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where L loc (T ; Θ) is the localization loss of ground truth as shown in (14) ; L obj (T ; Θ) is the object confidence loss of ground truth as shown in (15); L conf (T ; Θ) is the class confidence loss of ground truth as shown in (16); We use sum of the squared errors function f SSE to calculate the loss between the ground truth and the prediction, as shown in (13).…”

Section: Inputmentioning

confidence: 99%

Lightweight Deep Learning based Intelligent Edge Surveillance Techniques

Yu¹,

Yin²,

Gui³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Decentralized edge computing techniques have been attracted strongly attentions in many applications of intelligent internet of things (IIoT). Among these applications, intelligent edge surveillance (LEDS) techniques play a very important role to recognize object feature information automatically from surveillance video by virtue of edge computing together with image processing and computer vision. Traditional centralized surveillance techniques recognize objects at the cost of high latency, high cost and also require high occupied storage. In this paper, we propose a deep learning-based LEDS technique for a specific IIoT application. First, we introduce depthwise separable convolutional to build a lightweight neural network to reduce its computational cost. Second, we combine edge computing with cloud computing to reduce network traffic. Third, we apply the proposed LEDS technique into the practical construction site for the validation of a specific IIoT application. The detection speed of our proposed lightweight neural network reaches 16 frames per second in edge devices. After cloud server fine detection, the precision of the detection reaches 89\%. In addition, the operating cost at the edge device is only one-tenth of that of the centralized server.

show abstract

Section: Inputmentioning

confidence: 99%

Lightweight Deep Learning based Intelligent Edge Surveillance Techniques

Yu¹,

Yin²,

Gui³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…By reconstructing data with minimum error, SC searches for a set of over-complete (k n) bases whose sparse linear combination can approximately represent the data, with a penalty term (the second term in the cost function) restraining the combination coefficients to be sparse, which is the controlling weight in allocating the proportion of the reconstruction error and coefficient sparsity [16]. As more than one nonzero coefficient is allowed, sparse coding is better at precisely representing the data than K-means.…”

Section: General Models and Neural Response A K-means Sc And Icamentioning

confidence: 99%

“…Additionally, we propose a constraint here on intermediate representations, that features extracted in the hidden layer must be statistically independent, i.e., uncorrelated. It is motivated by the following qualitative reasoning: factors of variation are different aspects of the data which can vary separately and often independently; independent features can increase the dimension of observed vectors to obtain more information; a lot of evidence has shown specificity and selectivity of neurons in biological nervous systems [15], [16], [18]- [20]. An intuitive hallmark is that when we humans get stuck when attempting to recognize an object that is not readily identifiable, we tend to observe different parts of it to obtain more information to define it.…”

Section: Introductionmentioning

confidence: 99%

A Specific and Selective Neural Response Representation With Decorrelating Auto-Encoder

et al. 2019

View full text Add to dashboard Cite

Since the pioneering report with an unsupervised pre-training principle was published, deep architectures, as a simulation of primary cortexes, have been intensively studied and successfully utilized in solving some recognition tasks. Motivated by that, herein, we propose a decorrelating regularity on autoencoders, named decorrelating auto-encoder (DcA), which can be stacked to deep architectures, called the SDcA model. The learning algorithm is designed based on the principles of redundancy-reduction and the infomax, and a fine-tuning algorithm based on correlation detecting criteria. The property of our model is evaluated by auditory and handwriting recognition tasks with the TIMIT acoustic-phonetic continuous speech corpus and MNIST database. The results show that our model has a general advantage as compared with four existing models, especially in low levels, and when training samples are scarce our model put up stronger learning capacity and generalization. INDEX TERMS Machine learning, pattern recognition, deep architecture, auto-encoder, de-correlation.

show abstract

“…In the last few years, deep learning researches focused on both academia and industry [26] to process massive data with limited computing resources, extract high-dimensional features, and strong nonlinear fitting ability [27]- [29]. In civil engineering, DL based methods have been widely adopted in crack identification [30], [31], micro-seismic event detection and location in underground mines [32] as well as safety management in construction sites [33]- [35]. To learn more patterns from the past time series, the latest scholars started to use a more advanced type of new network, namely, the long short-term memory (LSTM) neural network.…”

Section: Introductionmentioning

confidence: 99%

Deep Learning Model for Shield Tunneling Advance Rate Prediction in Mixed Ground Condition Considering Past Operations

2020

View full text Add to dashboard Cite

The advance rate (AR) is a significant parameter in shield tunneling construction, which has a major impact on construction efficiency. From a practical perspective, it's helpful to establish a predictive model of the AR, which takes into account the instantaneous parameters as well as the past operations. However, for shield tunneling in mixed ground conditions, most researches focused on the average values of AR per ring and neglect the influence of past operations. This paper presents a long short-term memory (LSTM) recurrent neural network model, which was developed for the slurry shield tunneling in a mixed ground of round gravel and mudstone in Nanning metro. A temporal aggregated random forest is employed to rank the importance of the explanatory features. The model performances in different ground conditions are investigated. The results show that the LSTM model can be effectively implemented for the AR prediction. A high correlation is observed between predicted and measured AR with a correlation coefficient (R 2) of 0.93. The LSTM based AR predictive model is compared with the random forest (RF) model, the deep feedforward network (DFN) model, and the support vector regression (SVR) model. The comparison shows that the LSTM model has the best performances compared to other models. With onefourth features, we can achieve a 95% prediction accuracy measured by the R 2 in the proposed model.

show abstract

Deep Learning for Risk Detection and Trajectory Tracking at Construction Sites

Cited by 61 publications

References 51 publications

Lightweight Deep Learning based Intelligent Edge Surveillance Techniques

Lightweight Deep Learning based Intelligent Edge Surveillance Techniques

A Specific and Selective Neural Response Representation With Decorrelating Auto-Encoder

Deep Learning Model for Shield Tunneling Advance Rate Prediction in Mixed Ground Condition Considering Past Operations

Contact Info

Product

Resources

About