Table 8.

Literature result (continued, Part 3 of 5): summary table of papers reporting on objective, analysis methods, and obtained results.

Author(s)	Objective	Analysis Methods	Results
Li et al., 2019 [86]	Introduce the WisDriver system, which detects 15 different dangerous driving behaviours	Multiple approaches for signal processing (sliding window, mean absolute deviation): PCA, DTW, discrete wavelet transform (DWT)	CSI plus sensor can achieve up to 92% detection accuracy
Lindqvist and Hong, 2011 [87]	Conduct user interaction research to design driver-friendly smartphone applications that do not distract the driver	Interaction designs (no analysis)	Initial interaction designs for Android apps
Liu et al., 2017 [88]	Recognition of internal driver inputs (e.g., steering wheel angle, vehicle speed, and acceleration) and external perceptions of the road environment (e.g., road conditions and front view video)	Signal processing, filtering approaches, deep neural networks	Estimate steering wheel angle with an average error of 0.69, infer vehicle speed with an error of 0.65 km/h, and estimate binary road conditions with 95% accuracy
Ma et al., 2017 [89]	Propose a scheme to identify three dangerous driving behaviours, speeding, irregular change in direction and abnormal speed control	Coordinate reorientation, sensor error estimation, data correction, speed estimation, turn-signal identification	Kalman filter approach: average precision and recall for direction change and abnormal speed detection are 93.95% and 90.54%, respectively,
Mantouka et al., 2022 [91]	Identify unsafe driving styles and provide personalised driving recommendations	Two-stage K-means clustering	Summary statistics on collected trip data
Mantouka et al., 2019 [90]	Identify driver safety profiles from smartphone data and distinguish normal driving from unsafe driving	Unsupervised learning: two-stage K-means clustering approach	7.5% of the trips are characterised by distracted driving
Meiring et al., 2015 [92]	Review solutions and approaches to driving style analysis to identify relevant ML and AI algorithms	N/A	N/A
Meng et al., 2015 [93]	Develop a system that extends the driver’s view in all directions by using cameras from multiple cooperating smartphones in surrounding vehicles	Image processing	System detects a vehicle within 111 ± 60 ms
Mihai et al., 2015 [53]	Develop a system to determine the orientation of the driver’s head to infer visual attention	Image processing (OpenCV)	Feasibility tests in two scenarios, no numbers given
Nambi et al., 2018 [94]	Develop a windscreen-mounted, smartphone-based system to monitor driving behaviour (including driver states)	Android app: uses OpenCV, TensorFlow, and custom libraries (DNN and SVM)	Demonstration case, no further information provided by the authors
Omerustaoglu et al., 2020 [51]	Introduce a two-stage driver distraction detection system that integrates vehicle sensor data into a vision-based distraction detection model	CNN, LSTM-RNN on sensor and image data together; model tuning and transfer learning (from StateFarm to own dataset)	Increased overall accuracy to 85% compared to using only image data. Increased driver detection accuracy to 85% using sensor data.
Othman et al., 2022 [95]	Introduction of a driver state identification dataset synchronised with vehicle telemetry data	Dataset provision, unsupervised learning approach (K-means)	Clustered, labelled dataset
Pargal et al., 2022 [96]	Present an approach to detecting whether a smartphone is being used by the driver	Spectral analysis, power analysis of noise features, acoustic-based smartphone localisation	F1 scores from 0.75 to 0.875 for different smartphone placement scenarios