29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

Trajectory Prediction: A Review of Methods and Challenges

in Construction Safety

Chronopoulos C.1, Teizer J.2, Esterle L.3

1
Department of Civil and Architectural Engineering, Aarhus University, Denmark
2
Department of Civil and Mechanical Engineering, Technical University of Denmark, Denmark
3
Department of Electrical and Computer Engineering, Aarhus University, Denmark
[email protected]

Abstract. This review focuses on methods for trajectory prediction of moving entities (i.e.,
pedestrian workers and heavy construction equipment) in construction. To the authors’ knowledge
it is the first review on trajectory prediction devoted to construction safety. Through a bibliometric
analysis of the relevant literature, it examines the input data and prediction models used for trajectory
prediction in dynamic and complex construction environments. Several techniques are available to
perform prediction and their performance varies widely. Various types of data is being used,
however, so far vision-based data is the major input to the models. Hence, computer-vision
techniques are deployed for object tracking to infer the locations of the construction resources in
almost entirely outdoor environments. This review concludes with an overview of the gaps,
challenges, and future research steps for trajectory prediction relevant for researchers as well as
practitioners working on reducing occupational health and safety hazards on construction sites.

1. Introduction
Construction sites are highly dynamic and constantly evolving environments. Due to the
irregular environment, workers are prone to accidents mainly attributed to four key hazards,
namely falls from height, struck-by heavy objects, caught-in or –between, and electrocutions.
Furthermore, many construction accidents involve pedestrian workers in the proximity of static
or dynamic hazards, such as unprotected leading edges, moving heavy construction equipment,
or lifted crane loads. In 2019, nearly two-thirds of all construction fatal accidents in the U.S.
were caused by those types of hazards also known as the ‘Construction Focus Four’ (CPWR,
2021). This is a reason why construction in the US remains at the top of the list of fatal accidents
as it observes a 25%-share of all fatalities for the year 2019 (U.S. Bureau of Labor Statistics,
2020).
Trajectory prediction literature in the field of construction contributes mainly towards two
directions. First, is the development of proactive real-time safety systems based on proximity
monitoring for accident prevention. Those systems aim to provide relevant stakeholders (e.g.,
workers, equipment operators, safety managers) information for identifying static or dynamic
hazard zones and performing safety decision-making, as well as enough response time for
preventing imminent potentially hazardous events. Second, is the transition of construction to
automation and autonomy where trajectory prediction is critical for safety planning and
collision avoidance in human-robot collaboration.
Proximity monitoring and detection in construction sites have been a major research area, and
various location tracking methods have been adopted for acquiring spatio-temporal data of
onsite moving workers and equipment. For instance, radio frequency identification (RFID)
technology has been adopted in proximity warning systems (Schiffbauer, 2001; Teizer et al.,
2010), as well as ultra wideband (UWB) (Cheng et al., 2011; Teizer and Cheng, 2015),
Bluetooth low-energy (BLE) (Lin et al., 2015; Teizer et al., 2017), Global Navigation Satellite
Systems (GNSS) (Li et al., 2013; Zhang et al., 2015) and Long Range Wide Area Networks

https://doi.org/10.7146/aul.455.c233 428
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

(LoRA) (Teizer et al., 2020) for indoor and outdoor localization accordingly. More recently,
vision-based systems that take advantage of modern computer vision techniques have been
developed to identify the location of objects in construction site environments (Nahangi et al.,
2018; Kim et al., 2020). Those studies investigated and validated the applicability of such
location tracking technologies in various construction site environments and demonstrated high
accuracy. However, the uncertainties in the dynamic construction environment affect the
accuracy of proximity warning systems and alert frequency, reducing users’ trust and
potentially limiting their situational awareness (Ruff, 2006).
Regardless of when the construction sector will eventually achieve the envisioned level of
autonomy, both directions are significant to ensure and reinforce the safety of workers in the
dynamic and complex construction environment. The remainder of this paper is structured as
follows, (a) an introduction to the methodology of the literature review, (b) quantitative results
on the applications of trajectory prediction in construction as well as what input data and
prediction models are used, and (c) discussion of the challenges and future directions. The main
contributions of this paper are (i) identification of applications and methods for trajectory
prediction used in construction, (ii) limitations and challenges of the existing studies, and (iii)
future research steps in trajectory prediction for safety in construction.

2. Method
The literature review and the bibliometric analysis are based on the Scopus and Web of Science
(WoS) scientific databases because of the extensive coverage of literature, the ability to export
the search results as comma separated values (CSV) file for further analysis, and the support of
Boolean (i.e., “AND”, “OR”) and proximity operators (i.e., “W/”, “NEAR/”) in search strings
for advanced queries. The selected scientific databases have been successfully used in previous
state-of-the-art review papers (Jacobsen and Teizer, 2022; Kim et al., 2019). The yielded search
results have been exported in CSV file format and used for further analysis. The export contains
the citation information (e.g., authors, title, and year) and, abstract and keywords.
The literature search is performed as a keyword search query by using keywords and, Boolean
and proximity operators to limit the yielded results to the intended focus area. The search string
consists of two parts. The first part contains the keywords “trajector*”, “move*”, “path” and
“motion” followed by “predict*” and “forecast*” combined with the proximity operator “W/2”
(or “NEAR/2 in WoS)”, that is for combinations of sets of keywords within two words space.
The “or” operator is used to include all keyword sets variations. The asterisk wildcard is used
to indicate a character that may or may not be present in the term. The second part is focused
on yielding results that are relevant to the construction sector and therefore, it includes the
keyword “construction” followed by keywords such as, “worker”, “site”, “safety”, “project”
and “environment”. The results are limited to publications written in English.

3. Data
The publications that were yielded from the search were first filtered by eliminating the
duplicate records in the two databases. Subsequently, the results were screened to only include
relevant publications. The screening was done by searching in the titles and abstracts for topics
and research fields that are not in the scope of the review. These topics include for instance soil
mechanics and geotechnical engineering, offshore engineering, cost estimation, and labour
ergonomics, and therefore, the corresponding papers were excluded from the publications
database. This resulted in a total of 49 papers that were further assessed for eligibility.

429
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

The overall review process can be described by the following stages, (i) identification of the
records from the selected digital databases and duplicate elimination, (ii) screening of the initial
results, (iii) eligibility assessment of the publications, and (iv) inclusion of the relevant studies
(as shown in Figure 1). The analysis of the current state-of-the-art for the trajectory prediction
in construction aims to shed light on two main questions: What models or methods have been
used for trajectory prediction in construction in previous studies, and what type of input data
are utilized? For this, the abstracts and full text of the selected publications were reviewed.

Figure 1: Stages and flow of the review process (number of publications in parentheses).

3.1 Input Data

Each publication on the topic of trajectory prediction in construction has been categorized based
on the type of input data that are used by the corresponding predictive method to perform the
prediction task. Regardless of the technology used, trajectory prediction requires either
acquiring raw location tracking data (i.e., x, y, z coordinates over time) or inferring the location
of the tracked entities indirectly, for instance, through the analysis of video-recorded footage
and the application of object tracking techniques.
Therefore, the input data could be grouped into three main categories, (i) vision-based data, (ii)
raw location tracking data, and (iii) point clouds. In the analysed publications, the most
frequently used type of input is vision data, namely camera footage from construction sites. The
cameras used in the corresponding studies are mostly stationary and less frequently embedded
in moving vehicles, such as unmanned aerial vehicles (UAVs). Another type of input data used
is location tracking data from GNSS devices. Lastly, 3-dimensional point cloud data from Light
Detection and Ranging (LiDAR) sensors which are combined with kinematic data from inertial
motion units, stroke sensors, and rotational encoders that measure force, angular rate and
displacement of objects. Figure 2 shows the number of publications per type of input data.

3.2 Model
To understand how the trajectory prediction of moving workers and equipment is performed in
the construction safety literature, the adopted models and methods for trajectory prediction have
been investigated in the publications dataset. Deep neural networks (DNN) are commonly used
for trajectory prediction. A sub-genre of DNN is recurrent neural networks (RNN) in which
long short-term memory (LSTM) models are the most common for trajectory prediction

430
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

problems. In other publications, Kalman filtering (KF) and hidden Markov models (HMM) are
used for the prediction of workers and equipment motions. An illustration of the distribution of
adopted methods in the identified publications is presented in Figure 3.

Figure 2: Publications per year and type of input data used for trajectory prediction.

Figure 3: Publications per year and method used for trajectory prediction.

4. Results
Trajectory prediction is a critical topic in other relevant research fields, motivated and enhanced
by the rapid technological advancements of the computational power as well as of the
availability and cost-effectiveness of computational, sensory and data acquisition technologies.

4.1 Trajectory Prediction in Construction

Trajectory prediction in construction refers to the short-term prediction of the path followed by
a moving object within 1 to 10 seconds ahead and focuses on two main directions. First, the
development of proactive real-time safety systems based on proximity monitoring for accident
prevention (Golovina et al., 2019), and second, the transition of construction to automation and
autonomy where trajectory prediction is critical for safety planning and collision avoidance in
human-robot collaboration. Although the future of automation and robotics in construction is
promising (Garcia de Soto and Skibniewski, 2020), the majority of the identified publications
focus on proximity monitoring for accident prevention rather than on construction automation.

431
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

Three categories of input data were found to be used for trajectory prediction in construction
literature: vision-based data, raw location tracking data, and 3-dimensional point cloud data
from LiDAR sensors. Similar to Jacobsen and Teizer (2022) the average time of publication
(ATP) is used as a bibliographic metric. Table 1 shows the ATP for the different types of input
data, prediction methods and applications in the construction literature. The most common input
type, namely the vision-based data have an ATP of 2019.2, whereas the LiDAR data have the
highest ATP with very limited publications. DNN models are the most commonly used and
have the highest ATP contrary to the other predictive models that have also developed in the
previous decades. This is arguably showing a trend in the application of DNN for trajectory
prediction in construction. The three types of input data and models adopted for trajectory
prediction in the identified construction literature are depicted in Table 2 and further discussed
in the following sub-sections.

Table 1: Average time of publication (ATP) and number of papers for different types of input data
and model adopted for trajectory prediction in construction.
Input data ATP Publications Method ATP Publications
Vision 2019.2 11 DNN 2020.1 7
GNSS 2017.3 3 Kalman Filter 2019.0 5
LiDAR 2020.5 2 HMM 2017.3 3
Other 2016.6 1

Table 2: Model adopted for different types of input data for trajectory prediction in construction.

Input data DNN Kalman Filter HMM Other References

Vision x Rezazadeh Azar, (2016)
x (ConvLSTM) Bang et al., (2021)
x (Seq2Seq LSTM) Cai et al., (2020)
x Deng et al., (2021)
x (Seq2Seq LSTM) Hu et al., (2020)
x (Social LSTM+GAN) Kim et al., (2019)
x (Social LSTM+GAN) Kim et al., (2020)
x (Social LSTM) Kong et al., (2021)
x (UKF) Papaioannou et al., (2017)
x (LSTM+MDN) Tang et al., (2020)
x Zhu et al., (2016)
GNSS x Rashid and Behzadan, (2017)
x Rashid et al., (2017)
x Rashid and Behzadan, (2018)
LiDAR x (EKF) Rasul et al., (2020)
x (UKF) Rasul et al., (2021)

Vision-based Data
Video recorded footage is used for predicting the movement of workers and equipment in
construction sites through vision-based object recognition. The tracked objects (i.e., workers

432
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

and equipment) are identified in the frames using computer vision and the motion vector is then
calculated. Short-term prediction is commonly performed based on NN models and KF,
whereas HMM are less frequently applied. Zhu et al. (2016) proposed a framework for
computer vision-based estimation of position and short-term prediction of workers and mobile
equipment. The researchers assumed functionality with clear and of acceptable quality videos
with limited occlusions, which makes the framework susceptible to input of inferior quality. To
solve the tracking limitations in construction environments, Rezazaddeh Azar (2016) developed
a vision-based equipment tracking algorithm for automated camera control with predictive
capability by estimating the motion vector and speed of the tracked object.
To increase the accuracy of the predictive models semantic and contextual information is used
combining input from other sensory technologies. Papaioannou et al. (2017) introduced a
system that uses footage from CCTV camera infrastructure and data from the inertial sensors
embedded on modern smartphones and applied the Social Force Model (SFM) to consider
obstacles and other people in the scene, assuming that they affect the behaviour of human
motion, and represent their effect as repulsive forces. Cai et al. (2020) designed an LSTM model
to predict worker trajectories in construction environments, considering additional contextual
information, namely the distance to the nearest neighbour, the relationship between that
neighbour and the tracked worker, and the distance to destination. An LSTM network combined
with mixture density network (MDN) for construction workers and equipment path prediction
towards right time intervention of collision and intrusion was constructed by Tang et al. (2020).
The model considers two contextual cues, namely the distance between moving and static
objects and the type of objects (i.e., worker and vehicle) to predict up to 2 seconds in the future.
Although the model outperforms other existing models, it is still limited by the dynamic visual
occlusions due to other moving construction resources. Semantic information in the form of
predefined hazard zones is also considered in the literature. Deng et al. (2021) used KF to
predict the movement of workers in construction sites and the estimated trajectory is checked
against a set of artificial danger zone boundaries to determine whether the prediction point lies
inside or outside of the zones. Considering the occlusion limitations, the researchers performed
multi-angle detection which however, is limited by the camera resolution, especially when the
workers are far from the camera position. Kong et al. (2021) proposed a framework for workers’
trajectory prediction in construction sites based on the Social LSTM architecture. The
framework takes into consideration the workers’ unsafe behaviour, defined as any movement
towards predefined hazardous areas, and corrects the predicted trajectories using KF. One
important shortcoming of that study is related to the validation of the pre-trained model,
performed on their own dataset with limited scenarios, preventing it from being generalizable.
Only two of the identified publications focused on the future construction, where human
workers and robots co-exist and collaborate. Kim et al. (2019) proposed a framework based on
social generative adversarial network (S-GAN) for trajectory prediction to tackle contact-driven
hazards in construction between workers and autonomous trucks. Their results showed that
longer observation periods do not necessarily lead to higher prediction accuracy, due to
inclusion in the prediction of less relevant time steps. In a later study, they evaluated the model
on a controlled testbed, including a worker and a truck following three predefined movement
patterns (Kim et al., 2020). Hu et al. (2020) expanded the application of the LSTM model
developed by Cai et al. (2020), by implementing the A* path planning algorithm for
autonomous robots in construction sites. However, the study validates the worker trajectory and
path planning algorithms separately assuming flat ground surface.

433
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

Location Tracking Data

GNSS are satellite-based navigation technologies that depend on the satellites orbiting around
the earth. Existing studies have deployed low-cost GNSS technology for tracking construction
resources to enhance construction safety, planning and management (Pradhananga and Teizer,
2013; Zhang et al., 2015). GNSS data have also been used as input to trajectory prediction
models in construction applications. Rashid and Behzadan (2017) developed a smartphone-
based application for trajectory prediction of workers to prevent contact-driven accidents in
construction sites. The underlying model is based on HMM. A risk factor is introduced and
ranges between 0 and 1 depending on the angle between the trajectory and the centre of one
stationary and user-defined hazard zone (Rashid et al., 2017). The model was further developed
to consider one static or dynamic hazard (i.e., moving between two points) and validated it by
comparing to a benchmark Polynomial Regression model, showing better prediction accuracy
(Rashid and Behzadan, 2018). Both models however, are error-prone in predicting trajectories
with sharp turns and are limited to a single pedestrian worker and a predefined hazard.
Furthermore, the application considers outdoor construction activities due to the limitations of
GNSS technologies in indoor environments. Another shortcoming is related to the large number
of detected close-call events (n=369) and potential collisions (n=77) in a 30-minute experiment,
which could hinder the users’ situational awareness and trust in the warning system and lead to
delays.

Point Clouds
Point clouds are sets of data points in space that can represent 3-dimensional objects, where
each point has its own set of x, y and z coordinates. In a recent study, a LiDAR sensor was
utilized to acquire point cloud data to track the positions of heavy machinery and obstacles in
a construction site (Rasul et al., 2020; Rasul et al., 2021). The raw point cloud data were
analysed to first detect the heavy machinery (i.e., excavator) and then perform detection and
clustering of other objects (i.e., workers and machinery) of a width greater than 0.4m, which is
the average chest width of a human being. The Extended Kalman Filter (EKF) was adopted for
predicting the position and velocity of the moving objects, whereas the excavator’s predicted
working area was calculated based on kinematics analysis and data from embedded stroke
sensors and a rotational encoder (Rasul et al., 2020). In a later study, they used unscented
Kalman filtering (UKF) to predict the non-linear motion dynamics of the moving objects. In
both studies two safety indices are defined and used, namely the time to collision (TTC), and
the warning index (x) defined as the degree of potential collision risks.

5. Gaps, Challenges and Future Research

This paper reviews the problem of trajectory prediction in dynamic and complex construction
environments and discusses the current applications, prediction methods and input data used
for the predictive task. The following summarizes the gaps, challenges and directions for future
research for trajectory prediction with a specific focus on the construction site applications:
1. Predicting the future trajectories of moving objects while incorporating human behaviour
is a highly complex problem and hence it is limited to short-term prediction typically for 1-
10 seconds ahead with decreasing performance when the prediction horizon increases.
Previous studies improved the models’ accuracy by correcting the predicted trajectories
with KF, whereas other methods include construction semantic (i.e., static hazard zones)
and contextual information (e.g., distance from hazards, risk factor) to achieve better
performance. Trajectory prediction of moving construction resources should not neglect

434
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

pedestrian workers and their contribution to potentially hazardous events. Future research
should focus on both construction heavy equipment and pedestrian workers while
integrating additional construction semantic information. For instance, construction
semantic information such as dynamic hazard areas, work order and construction site layout
information (Chronopoulos et al., 2021) could potentially further improve the prediction
performance and thus, increase the impact of the developed architectures in real
construction applications.
2. Currently, trajectory prediction in construction aims to support the development of
proactive real-time safety systems for accident prevention. Increasing the accuracy of
prediction models is a common goal among all identified publications. However, it is
important to consider the time dimension in proactive warning systems and the feedback
must be shared at the desired level-of-detail at the right-time instead of in real-time (Teizer,
2016). Frequent warnings can limit the situational awareness and trust in the warning system
of workers and operators (Ruff, 2006). Novel systems focusing on construction safety
though trajectory prediction should ensure not only high prediction accuracy but also right-
time warning functionality.
3. The current state-of –the-art of trajectory prediction in construction mostly utilizes vision-
based data as input to computer-vision methods for object recognition and tracking. For
this, recorded videos from cameras on-board UAVs, stationary commercial cameras,
existing CCTV infrastructure and public datasets (Lerner et al., 2007) are used to train or
validate the developed models. This introduces three significant issues that need to be
considered in future research studies. First, is the comparability of performance metrics
between models that are validated on different datasets. Second, is the absence of publicly
available datasets for the construction sector to be used as benchmark for validating the
developed models. Third, the use of camera-based systems in dynamic and rough
construction environments is susceptible to various limitations, such as occlusions (Zhu et
al., 2016), camera equipment resolution (Deng et al., 2021) and limited field-of-view
(Jacobsen and Teizer, 2021), dust, weather conditions, malicious acts, and vandalisms.
4. The trajectory prediction literature in the context of construction focuses primarily on
outdoor applications and thus the proposed methods including the tools, models, training
and validation are structured based on that spatial assumption. However, several types of
private or public construction projects take place in indoor environments, for instance
buildings, underground and tunnelling projects with additional constraints (e.g., limited
luminosity, dust, no network, signal or GNSS coverage) that further constrain the
applicability and scalability of those methods in the aforementioned projects. Future
research efforts should also include indoor or hybrid construction environments in
developing and validating trajectory prediction models for construction safety.

6. Conclusion
This review paper is, to the authors’ knowledge, the first that presents an overview of the
trajectory prediction models for complex and dynamic construction environments. Deep neural
networks are commonly used in recent studies to perform prediction compared to other
methods. To increase the performance of the models, linear estimation models have been
applied and integration of limited contextual and semantic information is performed. Various
limitations exist and are related to the applied technologies as well as the dynamic and complex
characteristics of the construction environments. The paper discusses those limitations and
proposes future research steps in trajectory prediction for safety in construction. For instance,
the integration of additional construction semantic information (e.g., dynamic hazard areas,
work order and construction site layout information) to further improve the prediction accuracy

435
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

and the consideration of right-time proactive warning systems not only in outdoor construction
environments but also in indoor or hybrid construction projects.

References
Bang, S., Hong, Y. and Kim, H. (2021). Proactive proximity monitoring with instance segmentation and
unmanned aerial vehicle‐acquired video‐frame prediction, Computer-aided civil and infrastructure
engineering, 36(6), pp. 800–816. doi: https://www.doi.org/10.1111/mice.12672.
Cai, J., Zhang, Y., Yang, L., Cai, H. and Li, S. (2020). A context-augmented deep learning approach for
worker trajectory prediction on unstructured and dynamic construction sites, Advanced engineering
informatics, 46(101173), p. 101173. doi: https://www.doi.org/10.1016/j.aei.2020.101173.
Cheng, T., Venugopal, M., Teizer, J. and Vela, P.A. (2011). Performance evaluation of ultra wideband
technology for construction resource location tracking in harsh environments, Automation in
Construction, 20(8), pp. 1173–1184. doi: https://www.doi.org/10.1016/j.autcon.2011.05.001.
Chronopoulos, C., Johansen, K.W., Teizer, J., Schultz, C. and Esterle, L. (2021). Towards a holistic
self-organised safety framework for construction, in 2021 IEEE International Conference on Autonomic
Computing and Self-Organizing Systems Companion (ACSOS-C). IEEE. doi: https://doi.org/
10.1109/ACSOS-C52956.2021.00060.
CPWR - The Center for Construction Research and Training (2021). Fatal injury trends in the
construction industry. https://www.cpwr.com/wp-content/uploads/DataBulletin-February-2021.pdf.
Deng, H., Ou, Z. and Deng, Y. (2021). Multi-angle fusion-based safety status analysis of construction
workers, International journal of environmental research and public health, 18(22), p. 11815.
Garcia de Soto, B. and Skibniewski, M.J. (2020). Future of Robotics and Automation in Construction.
In Sawhney, A., Riley, M., and Irizarry, J. (eds.) Construction 4.0: An innovation platform for the built
environment. Routledge, pp. 289–306.
Golovina, O., Perschewski, M., Teizer, J., König, M. (2019). Algorithm for Quantitative Analysis of
Close Call Events and Personalized Feedback in Construction Safety, Automation in Construction,
Elsevier, 99, 206-222, doi: https://doi.org/10.1016/j.autcon.2018.11.014.
Hu, D., Li, S., Cai, J. and Hu, Y. (2020). Toward intelligent workplace: Prediction-enabled proactive
planning for human-robot coexistence on unstructured construction sites, in 2020 Winter Simulation
Conference (WSC). IEEE.
Jacobsen, E.L. and Teizer, J. (2022). Deep learning in construction: Review of applications and potential
avenues, Journal of computing in civil engineering, 36(2). doi: https://www.doi.org/
10.1061/(asce)cp.1943-5487.0001010.
Kim, D., Jebelli, H., Lee, S. and Kamat, V.R. (2020). Enhancing deep neural network-based trajectory
prediction: Fine-tuning and inherent movement-driven post-processing, in Construction Research
Congress 2020. Reston, VA: American Society of Civil Engineers.
Kim, D., Lee, S. and Kamat, V.R. (2020). “Proximity prediction of mobile objects to prevent contact-
driven accidents in co-robotic construction, Journal of computing in civil engineering, 34(4), p.
04020022. doi: https://www.doi.org/10.1061/(asce)cp.1943-5487.0000899.
Kim, D., Liu, M., Lee, S. and Kamat, V.R. (2019). Trajectory prediction of mobile construction
resources toward pro-active struck-by hazard detection, in Proceedings of the 36th International
Symposium on Automation and Robotics in Construction (ISARC). International Association for
Automation and Robotics in Construction (IAARC).
Kim, M.-K., Wang, Q. and Li, H. (2019). Non-contact sensing based geometric quality assessment of
buildings and civil structures: A review, Automation in Construction, 100, pp. 163–179.
Kong, T., Fang, W., Love, P.E.D., Luo, H., Xu, S. and Li, H. (2021). Computer vision and long short-
term memory: Learning to predict unsafe behaviour in construction, Advanced engineering informatics,
50(101400), p. 101400. doi: https://www.doi.org/10.1016/j.aei.2021.101400.

436
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

Lerner, A., Chrysanthou, Y. and Lischinski, D. (2007). Crowds by example, Computer graphics forum:
journal of the European Association for Computer Graphics, 26(3), pp. 655–664. doi:
https://www.doi.org/10.1111/j.1467-8659.2007.01089.x.
Li, H., Chan, G. and Skitmore, M. (2013). Integrating real time positioning systems to improve blind
lifting and loading crane operations, Construction management and economics, 31(6), pp. 596–605.
Lin, X.-Y., Ho, T.-W., Fang, C.-C., Yen, Z.-S., Yang, B.-J. and Lai, F. (2015). A mobile indoor
positioning system based on iBeacon technology, Annual International Conference of the IEEE
Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society.
Annual International Conference, 2015, pp. 4970–4973. doi: https://www.doi.org/10.1109/
EMBC.2015.7319507.
Nahangi, M., Heins, A., McCabe, B. and Schoellig, A. (2018). Automated localization of UAVs in GPS-
denied indoor construction environments using fiducial markers, in Proceedings of the 35th
International Symposium on Automation and Robotics in Construction (ISARC). International
Association for Automation and Robotics in Construction (IAARC). doi: https://www.doi.
org/10.22260/isarc2018/0012.
Papaioannou, S., Markham, A. and Trigoni, N. (2017). Tracking people in highly dynamic industrial
environments, IEEE transactions on mobile computing, 16(8), pp. 2351–2365.
Pradhananga, N. and Teizer, J. (2013). Automatic spatio-temporal analysis of construction site
equipment operations using GPS data, Automation in Construction, 29, pp. 107–122. doi:
https://www.doi.org/10.1016/j.autcon.2012.09.004.
Rashid, K.M. and Behzadan, A.H. (2017). Enhancing motion trajectory prediction for site safety by
incorporating attitude toward risk, in Computing in Civil Engineering 2017. Reston, VA: American
Society of Civil Engineers.
Rashid, K.M. and Behzadan, A.H. (2018). Risk behavior-based trajectory prediction for construction
site safety monitoring, Journal of construction engineering and management, 144(2), p. 04017106.
Rashid, K.M., Datta, S. and Behzadan, A.H. (2017). Coupling risk attitude and motion data mining in a
preemtive construction safety framework, in 2017 Winter Simulation Conference (WSC). IEEE.
Rasul, A., Seo, J. and Khajepour, A. (2021). Development of sensing algorithms for object tracking and
predictive safety evaluation of autonomous excavators, Applied sciences (Basel, Switzerland), 11(14),
p. 6366. doi: https://www.doi.org/10.3390/app11146366.
Rasul, A., Seo, J., Oh, K., Khajepour, A. and Reginald, N. (2020). Predicted safety algorithms for
autonomous excavators using a 3D LiDAR sensor, in 2020 IEEE International Systems Conference
(SysCon). IEEE.
Rezazadeh Azar, E. (2016). Active control of a pan-tilt-zoom camera for vision-based monitoring of
equipment in construction and surface mining jobsites, in Proceedings of the 33rd International
Symposium on Automation and Robotics in Construction (ISARC). International Association for
Automation and Robotics in Construction (IAARC).
Ruff, T. (2006). Evaluation of a radar-based proximity warning system for off-highway dump
trucks, Accident; analysis and prevention, 38(1), pp. 92–98. doi: https://www.doi.org/10.1016/j.aap.
2005.07.006.
Schiffbauer, W.H. (2001). An Active Proximity Warning System for Surface and Underground Mining
Applications, in SME Annual Meeting.
Tang, S., Golparvar-Fard, M., Naphade, M. and Gopalakrishna, M.M. (2020). Video-based motion
trajectory forecasting method for proactive construction safety monitoring systems, Journal of
computing in civil engineering, 34(6), p. 04020041. doi: https://www.doi.org/10.1061/(asce)cp.1943-
5487.0000923.
Teizer, J. (2016). The role of automation in right-time construction safety, in Proceedings of the 33rd
International Symposium on Automation and Robotics in Construction (ISARC). International
Association for Automation and Robotics in Construction (IAARC).

437
 29th International Workshop on Intelligent Computing in Engineering (EG-ICE)

Teizer, J., Allread, B.S., Fullerton, C.E. and Hinze, J. (2010). Autonomous pro-active real-time
construction worker and equipment operator proximity safety alert system, Automation in Construction,
19(5), pp. 630–640. doi: https://www.doi.org/10.1016/j.autcon.2010.02.009.
Teizer, J. and Cheng, T. (2015). Proximity hazard indicator for workers-on-foot near miss interactions
with construction equipment and geo-referenced hazard areas, Automation in Construction, 60, pp. 58–
73. doi: https://www.doi.org/10.1016/j.autcon.2015.09.003.
Teizer, J., Neve, H., Li, H., Wandahl, S., König, J., Ochner, B., König, M. and Lerche, J. (2020).
Construction resource efficiency improvement by Long Range Wide Area Network tracking and
monitoring, Automation in Construction, 116 (103245). doi: https://www.doi.org/10.1016/j.autcon.
2020.103245.
Teizer, J., Wolf, M., Golovina, O., Perschewski, M., Propach, M., Neges, M. and König, M. (2017).
Internet of things (IoT) for integrating environmental and localization data in building information
modeling (BIM), in Proceedings of the 34th International Symposium on Automation and Robotics in
Construction (ISARC). Tribun EU, s.r.o., Brno.
U.S. Bereau of Labor Statistics (2020). Number and rate of fatal work injuries, by industry sector.
Available at: https://www.bls.gov/charts/census-of-fatal-occupational-injuries/number-and-rate-of-
fatal-work-injuries-by-industry.htm (Accessed: January 10, 2022).
Zhang, S., Teizer, J., Pradhananga, N. and Eastman, C.M. (2015). Workforce location tracking to model,
visualize and analyze workspace requirements in building information models for construction safety
planning, Automation in Construction, 60, pp. 74–86. doi: https://www.doi.org/10.1016/j.autcon.
2015.09.009.
Zhu, Z., Park, M.-W., Koch, C., Soltani, M., Hammad, A. and Davari, K. (2016). Predicting movements
of onsite workers and mobile equipment for enhancing construction site safety, Automation in
Construction, 68, pp. 95–101. doi: https://www.doi.org/10.1016/j.autcon.2016.04.009.

438