AI-Supported Image Recognition Tools for Circular Furniture Flows: A Final Report for the Digital Decision Support for Refurbishment (2DSR) Project. Digital Decision Support for Refurbishment (2DSR) is a 3.5-year research project sponsored by Vinnova: Sweden’s Innovation Agency. This report describes two digital decision support tools developed as part of 2DSR by two furniture companies, NORNORM and OOAKI Living, each in collaboration with RISE. In the case of NORNORM the project aspired to develop an image recognition tool that could accurately classify the level of damage of articles of furniture. The project attempted to build a model using 3D renderings of NORNORM products as a reference against which to judge images of circulating furniture. Due to shortcomings with this approach, the project transitioned to using a commercially available large language model (LLM) plus prompt engineering and fine-tuning to classify furniture damage. We achieved an overall accuracy of approximately 60% but with high precision in key categories. In the OOAKI case, the project goal was to be able to identify sofa types and classify the width of a sofa into different range buckets using a single image taken by a lay customer. Results were satisfactory for sofa type identification and moderate for width estimation. While the project did not deliver workplace-ready tools, it produced functional prototypes and highlighted challenges and opportunities during a period of rapid technological and regulatory change in artificial intelligence.

Millions of vehicles are transported every year, tightly parked in vessels or boats. To reduce the risks of associated safety issues like fires, knowing the location of vehicles is essential, since different vehicles may need different mitigation measures, e.g. electric cars. This work is aimed at creating a solution based on a nano-drone that navigates across rows of parked vehicles and detects their license plates. We do so via a wall-following algorithm, and a CNN trained to detect license plates. All computations are done in real-time on the drone, which just sends position and detected images that allow the creation of a 2D map with the position of the plates. Our solution is capable of reading all plates across eight test cases (with several rows of plates, different drone speeds, or low light) by aggregation of measurements across several drone journeys. 

Emergency response system is a complex system of systems (SoS). The introduction of connected and autonomous vehicles (CAVs) introduces an extra dimension into the complexity. Future emergency response must be able to take into account of the autonomous vehicles with different automation levels and leverage the increasing connectivity and automation for efficient emergency response. Architecture frameworks have long been used for system engineering for large complex systems. The emerging unified architecture framework converges previous architecture frameworks for a unified one towards both military and civilian use. Based on the scenario of emergency response with CAVs on highways, this paper motivates an enterprise architecture for emergency response system of systems (ERSoS) with identification of the key challenges and opportunities in addition to a proposal of required capabilities. The work is a first iteration of an enterprise architecture for ERSoS with CAVs and forms part of the overall ERSoS architecture development process. 

Anonymization methods are one potential way of alleviating the risks of capturing personal information during data collections. The work presented here is based on one such method that, in turn, is based on generating images through machine learning to replace the original images. The chosen method merges both the original image and the generated one resulting in a risk of information from the original image leaking through to the final result. Here a possible approach to measure how much influence the original image has on the final product is presented .

Developing a machine learning-based vehicular safety system that is effective and generalizes well, capable of coping with all the different scenarios in real traffic is a challenge that requires large amounts of data. Especially visual data for when you want an autonomous vehicle to make decisions based on peoples’ possible intent revealed by the facial expression and eye gaze of nearby pedestrians. The problem with collecting this kind of data is the privacy issues and conflict with current laws like General Data Protection Regulation (GDPR). To deal with this problem we can anonymise faces with current identity and face swapping techniques. To evaluate the performance and interpretation of the anonymization process, there is a need for a metric to measure how well these faces are anonymized that takes identity leakage into consideration. To our knowledge, there is currently no such investigation for this problem. However, our method is based on current facial recognition methods and how recent face swapping work determines identity transfer performance. Our suggestion is to utilize state-of-the-art identity encoders like FaceNet and ArcFace to make use of the embedding vectors to measure anonymity. We provide qualitative results that show the applicability of publicly available identity encoders for measuring anonymity. We further strengthen the applicability of how these encoders behave on the VGGFace2 dataset compared to samples that have had their identity changed by Faceshifter, along with a survey regarding the anonymization procedure to pinpoint how strong facial anonymization is compared the vector distance measurements.

Despite the strong interests in creating data economy, automotive industries are creating data silos with each stakeholder maintaining its own data cloud. Federated learning (FL), designed for privacy-preserving collaborative Machine Learning (ML), offers a promising method that allows multiple stakeholders to share information through ML models without the exposure of raw data, thus natively protecting privacy. Motivated by the strong need for automotive collaboration and the advancement of FL, this paper investigates how FL could enable privacy-preserving information sharing for automotive industries. We first introduce the statuses and challenges for automotive data sharing, followed by a brief introduction to FL. We then present a comprehensive discussion on potential applications of federated learning to enable an automotive collaborative ecosystem. To illustrate the benefits, we apply FL for driver action classification and demonstrate the potential for collaborative machine learning without data sharing.

Traffic situations leading up to accidents have been shown to be greatly affected by human errors. To reduce

these errors, warning systems such as Driver Alert Control, Collision Warning and Lane Departure Warning

have been introduced. However, there is still room for improvement, both regarding the timing of when a

warning should be given as well as the time needed to detect a hazardous situation in advance. Two factors that

affect when a warning should be given are the environment and the actions of the driver. This study proposes

an artificial neural network-based approach consisting of a convolutional neural network and a recurrent neural

network with long short-term memory to detect and predict different actions of a driver inside a vehicle. The

network achieved an accuracy of 84% while predicting the actions of the driver in the next frame, and an

accuracy of 58% 20 frames ahead with a sampling rate of approximately 30 frames per second.