Ready-to-go Field Analysis

Summary

Field analysis methods will become more widespread in the future, particularly for monitoring athletes during their training. In the long term, these analysis methods could lead to significant improvements in training, whether in terms of physical preparation, injury prevention or optimisation of the sporting gesture. This perspective can be transposed to physical activity in the workplace for injury prevention, or for monitoring patients on a daily basis. However, the accuracy and quantity of the data available mean that this analysis is very partial at present, and requires the development of methods based on learning and biomechanics to increase the data available and maximise the relevance of the approach for the practitioner. The latter also requires the tools to be designed from a field perspective, by co-constructing the analysis tools with those practitioners.

Long term scientific objective

This research stream is one of the clearly stated application outcomes of the research team: to be able to provide biomechanical analysis tools that are accurate, quick to implement and easy to deploy with in-field occupational, clinical and performance applications. This objective requires direct contact with the actors of the field in order to understand and scientifically translate their problems, i.e. to resolve the scientific questions posed by the practical issues of support in the field. In the field of sport, the aim is to develop indicators that are representative of the activity in order to understand it better and identify the driving forces behind performance or the potential risks of injury and the resulting physical conditioning. For disability, the aim is to understand the evolution of a pathology through the quantification of activity in order to support the patient and his/her therapist in the physical management of this pathology. In the field of ergonomics, the aim is to provide biomechanical quantities that can be aggregated into representative scores to assess a worker’s level of exposure to the physical risk factors represented by his or her activity at work. In all these fields, the need to be in the field reflects the importance of the ecological validity of the assessment, which is only effective if it is measured as closely as possible to the actual conditions in which it is performed, using a set of sensors that is as minimally invasive as possible without reducing the quality of the proposed analysis.

Figure 1: From top to bottom, left to right: a) Comparison of centre of pression (CoP) mediolateral position between insole and force plate measurement after calibration during a walk, b) Schematic representation of the forces measured by an instrumented football shoe and a force platform, and their associated reference frames, c) Estimating joint torques at work, d) Coordinate systems and the rigid transformations between them during a cycling calibration task

Short term goals and actions

We are currently working on the following goals:

Imu based solutions to monitor sport performance: As regards to IMUs, one of the major issues in 3D joint kinematics assessment lies in the misalignment of sensor axes with the anatomical body segment axes which requires specific sensors-to-segment calibration methods. In this aim we previously developed a Novel Sensor-to-Segment Calibration Procedure Based on pedaling motion for 3D joint motion during cycling. Based on this expertise, we will go further by implementing new calibration methods and biomechanical models to other ecological contexts. In the continuity of the PPR Project NePTUNE, we will implement a subject specific 3D model based on IMU data in the CusToM library. Such developments will serve as a decision support tool for coaches by analyzing the changes in the hydrodynamic profile of elite swimmers over competitive seasons. In the same line, the MUSSAKA project will aim at designing a complete monitoring system for kayakists performance enhancement. Based on multiple sensors (e.g., strain gauges and IMU embedded into paddles, strain gauges at foot/boat interface, IMUs at each upper segment of kayakist. . . ), this monitoring system will be able to deliver in real-time adequate information to riverside, such that a trainer can follow and give advices to athletes during a whole training session. As previously raised, another challenge lies in the evaluation of more global kinematic parameters (acceleration, jerk, movement phases, cadences, etc.) as well as the analysis of multivariate time series, enabling the automatic classification of activity and the identification of biomechanical characteristics over long durations in ecological situations. We recently contributed to the development of an automatic swimming Activity recognition and lap time assessment method using a deep Learning approach [Del+22]. Others projects will aim at providing novel methods for variabilities determination in swimming using embedded sensors (e.g., data-driven approach to identify technical stroke profiles during swimming based on kinematical clustering of intra and inter-cyclic variability).
Estimating internal forces from an RGB stream: Within the frame of the cifre grant with the society Moovency, we seek at developing a way to estimate internal forces being involved in the realization of load carrying tasks. It asks to develop a deep learning approach able to estimate the pose of the subject and the contact of his hands and feets with the ground or an object, then to follow a classical motion analysis pipeline to estimate the internal forces. Such an approach is really appealing for ergonomics assessments since it enables from a simple RGB stream to estimate both posture and forces.
Novel view synthesis from sparse RGB inputs: novel view synthesis with implicit representation of the scene is a very active field of research in computer vision. This approach general requires training a specific neural network architecture for each new scene, and needs a large amount of view points as inputs. We explore how to use sparse RGB inputs, and ideally a unique point of view, to simulate new views. This is an important contribution for scenes where it is almost impossible to place several cameras, such as sports competition or manufactures. Generating these novel views from a unique RGB video opens new research and application avenues in human motion analysis “on the field”.
Better understand fatigue to enhance exercise performance: endurance exercise performance can be predicted to a high degree using just three parameters: maximal and fractional oxygen consumption, and exercise economy (efficiency). However, these are all affected by fatigue. To better understand how fatigue alters these parameters, we are focusing on two main factors: neuromuscular function as it determines fatigue resilience (durability) to a high degree, and musculo-skeletal stiffness as it contributes heavily to movement economy and performance. Beyond sports performance applications, this research is sought after in sports ergonomics such as in the footwear industry (carbon-plated shoes, cushioning effects) but also ergometer design, compression garments, bike geometry….
False start detection in sprinting: the current rule for false-starts used in official athletics competitions dates back to the 90’s and uses a two-step process: first, an alarm sounds if a force >25 kg is applied to the starting blocks within a 100 ms period after the gun start, and second, a visual assessment of total body motion by a referee if that force was accompanied by an actual “starting” motion. This confusing rule leads to several shortcomings (perceived unfairness, scientific basis of the 100 ms, differences between equipment manufacturers, influence of subjective judgement, …). Based on a collaboration with the FASST project from the University of Limerick (Ireland) led by Emeritus Pr Drew Harrison, we are developing a proof-of-concept overcoming these limitations using bespoke hand platforms and false-start detection algorithms to offer the international federation World Athletics with developing a scientifically-proven, streamlined process.
Force-Velocity Profiling for sports performance : Force-Velocity (F-V) profiling is an important issue for the dynamic analysis of sports in which acceleration is a crucial factor in the seek for performance, such as athletics, speed skating, rugby or football [Cro+18]; [Lah+20]. It can be used to assess force generation capacities as a function of different execution speeds and thus provides guidance for training, particularly with a view to optimising strength and conditioning. Generally, an electromechanical system enabling to pull and brake the subject (e.g. 1080 sprint – 1080 Motion, Vasteras, Sweden) is used to establish F-V profiling in a reliable and reproducible way. More specifically, in swimming, this profiling can be used to identify the dynamic parameters of propulsion [Gon+20] as well as the swimmer’s active drag [GO22]. As part of the Neptune project, a number of metrological projects are performed to develop methods for identifying a load to be applied to a swimmer with a view to accurately determining active resistance.
Analysing serve biomechanics from a 3D markerless system: Within the frame of a PhD grant from EUR DIGISPORT, we seek at validating a 3D markerless motion capture system to evaluate serve kinematics and kinetics during real tennis matches.To this end, various markerless methods such as OpenCap, Pose2Sim, DeepLabCut and Theia Markerless will be tested and compared with the gold standard of 3D marker-based motion capture. Markerless methods track points of interest (keypoints) placed on the participant’s body and the racket. Validating a 3D markerless system will be very helpful for coaches. For them, it is fundamental to evaluate sport motion “in situ” in competitive conditions where fatigue or stress can influence performance and the risk of injury which is not the case in laboratory situations.