ExoNet Database: Wearable Camera Images of Human Locomotion Environments

Neural architecture search has proven to be highly effective in the design of efficient convolutional neural networks that are better suited for mobile deployment than hand-designed networks. Hypothesizing that neural architecture search holds great potential for human pose estimation, we explore the application of neuroevolution, a form of neural architecture search inspired by biological evolution, in the design of 2D human pose networks for the first time. Additionally, we propose a new weight transfer scheme that enables us to accelerate neuroevolution in a flexible manner. Our method produces network designs that are more efficient and more accurate than state-of-the-art hand-designed networks. In fact, the generated networks process images at higher resolutions using less computation than previous hand-designed networks at lower resolutions, allowing us to push the boundaries of 2D human pose estimation. Our base network designed via neuroevolution, which we refer to as EvoPose2D-S, achieves comparable accuracy to SimpleBaseline while being 50% faster and $12.7\times $ smaller in terms of file size. Our largest network, EvoPose2D-L, achieves new state-of-the-art accuracy on the Microsoft COCO Keypoints benchmark, is $4.3\times $ smaller than its nearest competitor, and has similar inference speed. The code is publicly available at https://github.com/wmcnally/evopose2d .

Advances in golf club performance are typically based on the notion that golfer biomechanics do not change when modifications to the golf club are made. The purpose of this work was to develop a full-swing, forward dynamic golf drive model capable of providing deeper understanding of the interaction between golfer biomechanics and the physical properties of golf clubs. A three-dimensional biomechanical model of a golfer, a Rayleigh beam model of a flexible club, an impact model based on volumetric contact, and a spin-rate dependent aerodynamic ball flight model are used to simulate a golf drive. The six degree-of-freedom biomechanical model features a two degree-of-freedom shoulder joint and a pelvis to model the X-factor. It is driven by parametric joint torque generators designed to mimic muscle torque production, which are scaled by an eccentric-concentric torque-velocity function. Passive resistive torque profiles fit to experimental data are applied to the joints, representing the resistance caused by ligaments and soft tissues near the joint limits. Using a custom optimization routine combining genetic and search-based algorithms, the biomechanical golf swing model was optimized by maximizing carry distance. Comparing the simulated grip kinematics to a golf swing motion capture experiment, the biomechanical model effectively reproduced the motion of an elite golf swing.

A long-held assumption in golf research is that the driver-ball impact is accurately modelled as a collision between two free bodies, i.e., the clubhead is not attached to the shaft. The purpose of this work was to examine the validity of this assumption using multibody simulation and motion capture technology. Ten elite golfers were recruited to participate in a motion capture experiment to validate a Rayleigh beam model of a flexible club. Using the six degree-of-freedom motion of the grip as an input to the model, the simulated shaft deflections showed good agreement with the experiment. An impact model based on volumetric contact was integrated with the flexible club model and was used to compare the launch conditions of free-body and full-club impacts. Analysis of the launch conditions revealed that the shaft creates a stiffening effect that resists clubhead rotation during contact, corresponding to an increase in ball speed and suppression of the gear-effect relative to free-body impacts. The results demonstrate that shaft dynamics cannot be treated as negligible when evaluating driver impact mechanics.