A5016391896- Autonomous Vehicle Technology and Safety
- Anomaly Detection Techniques and Applications
- Traffic and Road Safety
- Advanced Neural Network Applications
- Human-Automation Interaction and Safety
- Explainable Artificial Intelligence (XAI)
- Traffic Prediction and Management Techniques
- Human Pose and Action Recognition
University of Skövde
2022-2024
Volvo Cars (Sweden)
2024
Volvo (Sweden)
2023-2024
Every year worldwide more than one million people die and a further 50 are injured in traffic accidents. Therefore, the development of car safety features that actively support driver preventing accidents, is utmost importance to reduce number injuries fatalities. However, establish this it necessary advanced assistance system (ADAS) understands drivers intended behavior advance. The growing variety sensors available for vehicles together with improved computer vision techniques, hence led...
Driver intention recognition studies increasingly rely on deep neural networks. Deep networks have achieved top performance for many different tasks, but it is not a common practice to explicitly analyse the complexity and of network's architecture. Therefore, this paper applies architecture search investigate effects network real-world safety critical application with limited computational capabilities. We explore pre-defined space three layer types that are capable handle sequential data...
Traffic fatalities remain among the leading death causes worldwide. To reduce this figure, car safety is listed as one of most important factors. actively support human drivers, it essential for advanced driving assistance systems to be able recognize driver's actions and intentions. Prior studies have demonstrated various approaches intentions based on in-cabin external video footage. Given performance self-supervised pre-trained (SSVP) Video Masked Autoencoders (VMAEs) multiple action...
Driver intention recognition (DIR) methods mostly rely on deep neural networks (DNNs). To use DNNs in a safety-critical real-world environment it is essential to quantify how confident the model about produced predictions. Therefore, this study evaluates performance and calibration of temporal convolutional network (TCN) for multiple probabilistic learning (PDL) (Bayes-by-Backprop, Monte-Carlo dropout, Deep ensembles, Stochastic Weight averaging - Gaussian, Multi SWA-G, cyclic Gradient...
Driver intention recognition studies increasingly rely on deep neural networks.Deep networks have achieved top performance for many different tasks, but it is not a common practice to explicitly analyse the complexity and of network's architecture.Therefore, this paper applies architecture search investigate effects network real-world safety critical application with limited computational capabilities.We explore pre-defined space three layer types that are capable handle sequential data (a...
Real-world applications of artificial intelligence that can potentially harm human beings should be able to express uncertainty about the made predictions.Probabilistic deep learning (DL) methods (e.g., variational inference [VI], VI last layer [VI-LL], Monte-Carlo [MC] dropout, stochastic weight averaging -Gaussian [SWA-G], and ensembles) produce a predictive but require expensive MC sampling techniques.Therefore, we evaluated if probabilistic DL are uncertain when making incorrect...