As both a mode of transport and recreational activity, cycling has well-known health, environmental, and economic benefits. For these reasons, it has been encouraged in many countries, including the Republic of Ireland. However, with increasing popularity, there have been concurrent increases in road traffic-related cyclist injuries. This project aims to characterise cyclist collisions, which are known to be underreported in police statistics, and develop tools to inform engineering interventions.

For data collection, a survey addressing collisions was distributed to cyclists across the country in 2018. Police investigation files were also analysed for a sample of fatal and serious cyclist collisions in Ireland to supplement the analysis with further details specific to these more severe cases. Logistic regression modelling was used to clarify biasing factors for police reporting for collisions, and taxonomy schemes were developed for cyclist-motorised vehicle and single cyclist collisions to ascertain the distribution of pre-crash manoeuvres, and resulting impact configurations.

Findings from chapter 3 indicate that the largest proportion of collisions was between cyclists and motorised vehicles (56%), followed by single cyclist collisions (29%), collisions with other cyclists (8%), and pedestrians (7%). The odds of police reporting for collisions with motorised vehicles was 20 times greater than single cyclist collisions, 10 times greater than cyclist-cyclist collisions, and 4 times greater than collisions with pedestrians. The odds of police reporting for serious injury collisions was 7 times greater than minor injury collisions. Furthermore, findings from chapter 4 highlights the relative importance of collisions resulting from the cyclist and vehicle travelling in the same direction, specifically, nearsidevi hook, vehicle lane changing, and overtaking manoeuvres are emphasised. Furthermore, cases involving the cyclist being struck from the side by vehicle fronts comprise a smaller share than in previous studies. Specifically, side to side impacts, impacts between the front of the cyclist/bicycle and the side of the vehicle, and impacts with open(ing) doors emerge as important impact configurations with the inclusion of self-reported cases. For single cyclist collisions, the importance of loss of traction of the tires due to slippery road conditions and interactions with tram tracks and kerbs are emphasised. In chapter 5 a number of common scenarios have been identified for fatal and serious cases from police investigation files. In particular, this study highlights the prevalence of left/nearside-turning Heavy Goods Vehicles (HGVs) at junctions in urban environments, and overtaking manoeuvres from bonnet-type vehicles in rural environments. Furthermore, findings indicate that single cyclist collisions are common. Future safety interventions should aim to address these collision types.

Single cyclist collisions are common and underreported in official statistics (chapters 3, and 4). Internationally in urban environments, light rail tram tracks are a frequent factor in these cases, however, they have not yet been the subject of engineering analysis. The prevalence of traffic camera footage in urban environments presents an opportunity for detailed sitespecific safety insights. Therefore in chapter 6 a video-based analysis is presented with an investigation into the frequency and risk of unsuccessful crossings on tram tracks in wet road conditions at 9 locations around Dublin city centre. A predictive model for crossing success is also devised as a function of crossing angle for use in a Surrogate Measure of Safety (SMoS) framework. Modelling results show that crossing angle is a strong predictor of crossing success, and that cyclist velocity is not. Findings indicate that infrastructural planners should design for cyclist crossing angles of 30° or greater. The prevalence of external factors which limit crossing angles for cyclists are highlighted. In particular, kerbs are a common factor, along with passing/approaching vehicles or other cyclists. Furthermore, a new SMoS for cyclist interactions with tram tracks is introduced, and its utility is demonstrated through an open-source application (SafeCross).

Findings from chapters 3, and 4 indicate that single cyclist falls are common. Furthermore, self-reported injury patterns in chapter 4 suggest that protective movements of falling cyclists may influence kinematics and dynamics. However, previous computational modelling attempts have not considered kinematic or dynamic components of protective response in single cyclist falls. Novel computer vision/deep learning-based 3D human pose estimation methods localise human joints from images and videos. Pose representation is normally limited to 3D joint positional/translational degrees of freedom (3DOFs), however, an additional three rotational DOFs (6DOFs) are required for many potential biomechanical applications. Positional DOFs are insufficient to analytically solve for joint rotational DOFs in a 3D human skeletal model.

Therefore, in chapter 7 a temporal inverse kinematics (IK) optimisation technique is proposed to infer joint orientations throughout a biomechanically informed, and subjectspecific kinematic chain. For this, link directions from a position-based 3D pose estimate are prescribed. Sequential least squares quadratic programming is used to solve a minimisation problem that involves both frame-based pose terms, and a temporal term. With this, in chapter 8 a novel pipeline is presented for reconstruction of falls from monocular footage. The IK algorithm is adapted for use with the MADYMO (MAthematical DYnamic MOdels) ellipsoid computational human body model, to initialise a model using 3D human kinematics obtained using a deep learning-based 3D human pose estimator. The method is applied to a case study of a sideways falling cyclist after an unsuccessful crossing of railway tracks. Findings from this study indicate that initialised pose and velocity significantly influence resulting kinematics, and that the limbs are used to prevent falls, or break falls by absorbing impact energy and protecting the head. This approach can be applied to a variety of impact biomechanics tasks not limited to cyclist falls.