Projets novateurs réalisés

Using RTLS and Computer Vision to Extend Worksite Safety

The project aims to extend worksite safety of construction projects at Hydro-Quebec (HQ) using computer vision and a Real-Time Location System (RTLS). The case study is a substation construction project near Montreal. The main safety risks that will be targeted in the case study are related to equipment mobility (struck-by accidents) and not wearing Personal Protection Equipment. The concept of the method is to have a priori information about the types of expected risks in the planning phase, and then to monitor the site using video cameras and the RTLS. Artificial Intelligence (AI) and computer vision techniques are used to detect the location and other attributes of the workers and equipment with respect to the identified risks. The workers will be equipped with a wristband that can generate vibration safety alerts in case of proximity to equipment. In addition to safety support, the system can provide the following side benefits: (1) improved security by detecting potential intruders to the construction site by using infrared cameras and night vision; and (2) generating time-lapse video of the project.

Deep learning based approaches for hard and soft data fusion towards better maritime domain awareness

In this project, we apply deep learning methods to analyze and obtain useful information from text data that are collected from social media, and combine these information with numerical data from physical sensors. We then develop new deep learning based solutions that exploit the combined data in order to track the ships in the open sea with more accuracy. The primary strength of our work is that social media data provides additional information when the usual physical sensors like radars and satellites can not provide enough data. Our work is integrated into the partner organization’s commercial software to illustrate the improved performance of ship tracking.

NetRepAIr: Making networks reliable for next-generation applications using AI/ML techniques

Networks have grown from small topologies connecting a dozen of devices to large, shared infrastructures supporting primary needs of our society. Today, we count on networked services for trading, commuting, monitoring weather conditions, meeting people. In order to provide reliable services, network operators need to cope with the daunting challenge of ensuring millions of flows from heterogeneous devices arrive at their destination on time and showing a reasonable throughput. Despite the significant advances recent Software Defined Networks (SDNs) provided towards managing large scale network infrastructures, they still fall short to enable fault-tolerant, performance-guaranteed data transmissions to the level next-generation applications such as 5G, smart cities, augmented reality and the Tactile Internet demand. In this project, we propose a new view to the problem of network reliability. Through Artificial Intelligence (AI) and Machine Learning (ML) techniques, we look for building a smart, highly scalable and robust network repair system. Our design will combine state-of-the-art machine learning techniques such as deep reinforcement learning and graph neural networks with high-performance and flexible network devices (e.g., P4 switches, NetFPGAs, and SmartNICs) to detect and correct network faults with high accuracy and in extremely short timescales.

Design of foot orthosis with customized variable stiffness structure using 3D printing techniques

There is a large need for custom orthotic insoles that meet user’s needs in terms of comfort, pressure distribution correction and impact absorption integrated a in more extensive and flexible way. The limited reliable control of these factors in current manufacturing processes has led to frequent orthotic adjustments, reduced device compliance, lowered effectiveness and increased time and expense from both the orthotic provider and the patient. Using 3D printing technology, this proposal aims to develop the next generation of custom orthotic insoles through tuning the mechanical characteristics of the insole within its structure. Upon project completion, results will lay a framework for designing optimized custom orthotic insoles.

Where do we want to go? Have we arrived? Improving transparency, rigourand knowledge in complex multi-stakeholder planning processes

This research aims to address to knowledge gaps in complex multi-stakeholder planning

contexts. First, it involves better understanding engagement methods and value elicitation

techniques with the purpose of decision making in multiple urban/rural planning contexts.

Second, it will research and develop a participatory monitoring and evaluation (M&E)

framework to be applied in a First Nations context, specifically to address the lack of

knowledge and application of M&E in the ‘Comprehensive Community Plans’ that are been

developed across the province supported by Indian and Northern Affairs Canada (INAC).

This research project is of relevance to the partner since it will greatly enhance the current

planning approaches and methods it is using. It will also deliver higher impacts for the clients

of the partner firm and hence be of benefit to Canada. Finally it has the potential to positively

influence a broader professional planning audience, particularly First Nations.

Multivariable PID Controller for Search and Rescue UAV Operations Based on Static Output Feedback

The research proposed in this document will build upon and extend the previously funded CRIAQ (AUT-1701) and MITACS (IT12130) projects on the development of a UAV platform for search and rescue activities in the ski facilities of Domain Saint Bernard in Mont Tremblant in collaboration with SII Canada. The goal of this research is to develop a synthesis methodology for a multivariable PID flight controller to steer a rescuing UAV to a person in danger using output feedback.

IoT device fingerprinting and anomaly detection using ML

The number of Internet of Things (IoT) devices is expected to reach 50 billion devices by 2020 and the devices are increasingly diverse. They are disrupting traditional security measures. Mobile Network Operators (MNOs) have limited control over customers’ IoT devices, as they are deployed on the customer premises. MNOs need to deploy effective security controls at their end to protect their assets. Huge amounts of data are generated by IoT devices, which can be exploited to understand device behaviours. The proposed research program aims at finding novel solutions to the problem of detecting abnormal behaviour in IoT environments. When abnormal network traffic is detected, two solutions can be adopted: blocking the traffic, or sending it for deeper analysis. The first solution may disconnect legitimate IoT devices, as certain behavior deviations are quite normal, e.g., bandwidth fluctuation. The second solution attempts to learn more about IoT devices and refine the learned behaviour model. This is a real-time and continuous learning process that adapts the model to a changing environment, e.g., new device types. Therefore, sophisticated IoT fingerprinting exploiting machine learning algorithms is the ultimate objective to achieve.

Creative Artificial Intelligence in Interactive Mobile Systems

The large amount of information available today on the web brings many challenges to the information retrieval and artificial Intelligence communities. Moreover, personalization is a key component in today’s successful mobile websites and interactive applications. In order to be effective, these websites are required to provide visitors with the information they need without the complexity in finding it. Developing intelligent and interactive systems with visual user interfaces is therefore essential for any mobile device. In this regard, our goal is to develop a new interactive mobile tool that elicit and lean users’ requirements and preferences, in order to provide them with what they actually need. This will be achieved by taking advantage of the advancement of artificial intelligence as well as the new 5G technologies.

Anomaly detection using AI/ML for Network Correction

Anomaly detection or outlier detection is a technique to identify rare items, observations or events which are differing significantly from most of the data or do not conform to the expected behavior of the system. Typically, anomalous data cause numerous problems in the computer networking and communication system. This project aims to develop an advanced anomaly detection algorithm by utilizing state-of-the-art machine learning and artificial intelligence techniques and combining it with existing anomaly detection techniques. We propose to develop a unique deep learning methodology based on the Modified Support Vector Machine (MSVM) and the Bi-directional Long Short-Term Memory Recurrent Neural Networks (BLSTM RNN) approaches. We will test and evaluate the solution with respect to the accuracy, miscalculation rate, precision, true positive rate and F1 score

AI-Blockchain integration to ensure explainability, accountability, traceability and reproducibility (EATR) in mission critical systems

This project is an effort to integrate two of the emerging technologies of our time, i.e., Artificial Intelligence and Blockchain. The purpose of this integration is to overcome the drawbacks of individual technologies and work in coherence for mutual benefits. Most of the current AI systems currently deployed are black boxes and they do not provide explanations of suggested decisions. With the integration of blockchain with AI, it is possible to create a transparent AI system with trail of data processing to address the ‘Why?’ question. Moreover, in the era of Internet of things, enforcement of GDPR and 5G in sight, it is compulsory for organizations to ensure the data and identity protection of the processes and people utilizing them. The objective of this research is to develop a logic model that complies with the privacy requirements as well as being explainable in its suggested course of action. The possible applications of such systems are in financial and IT compliance/ responsibilities/penalties insurance, governance and security. To that end, a system can be designed integration AI and blockchain technologies to continuously monitor different processes and ensure that all the functionalities are in place and working properly.

Ensemble-based Dimensionality Reduction Model for Wireless Time-Series

The new wireless network technology will provide users with a higher communication quality. However, we will face two critical problems: the wireless traffic will increase considerably, and the wireless signals will contain noise. The Wifi signals are represented as time series, but processing and removing noise from such huge-volume, high-dimensional and complex data pose great challenges. Learning from time-series is an ambitious problem, and the curse of dimensionality makes machine learning algorithms incompetent.

Dimensionality Reduction Algorithms (DRAs) can effectively address the problems above. Still, the DRA application to time series has been limited due to data complexity. We will first examine the characteristics of the wireless data and investigate which DRAs are the most suited. Next, we will define strategies to combine DRAs, a challenging task but significant to improve the DRA accuracy. By conducting an empirical analysis, we will develop the ensemble DRA that best maximizes the network performance.

Crowdsensing-based Wireless Indoor Localization using an Innovative AI & ML Algorithm

Smartphone based indoor navigation services are desperately needed in an indoor GPS-denied environment, such as in Combat-zone Surveillance, Health Monitoring, Fire Detection, etc. The Receive Signal Strength (RSS) based algorithms are commonly used in indoor localization, which rely on the WiFi fingerprint data built by the Mobile Crowdsensing approach. Conventional statistical and probability techniques are used to deal with crowdsensing-based RRS fingerprinting information, but there are some issues such as low localization accuracy, highly relevance to the device/software they used, large database required, unfriendly to new encountered smartphone, etc. The proposed project will develop a novel integrated approach that combines AI technology (e.g., Artificial Neural Networks) and ML methods (e.g., Feed-forward Multilayer Perceptron Regressor and Support Vector Machine) to solve aforementioned problems, as well as build a crowdsensing-based RRS Wi-Fi fingerprinting dataset in a university building in Regina, SK Canada for indoor positioning studies