Independent Study: Jake

Investigation of methods and existing solutions to detect Lower-limb Asymmetry using Computer Vision

Introduction

Lower-limb asymmetry is an important indicator in physiotherapy and rehabilitation, as differences between the left and right lower limbs may reflect muscle weakness, reduced mobility, or incomplete recovery after injury. With recent advances in computer vision and pose estimation models such as YOLO Pose and MediaPipe, it is now possible to estimate joint angles, movement patterns, and limb symmetry using only RGB cameras, without the need for wearable sensors or motion capture systems. This independent study investigates how purely computer vision-based methods can be used to detect and evaluate lower-limb asymmetry.

Existing Methods

Traditional methods for assessing lower-limb asymmetry in physiotherapy mainly rely on observation, joint range of motion evaluation, gait analysis, force plates, wearable sensors, and motion capture systems. Although these methods can provide accurate measurements of movement quality and body balance, many of them require expensive equipment, calibration procedures, or sensors attached to the patient’s body.

Recent developments in computer vision, depth cameras, and RGB pose estimation models such as YOLO Pose have introduced a lower-cost and more accessible alternative for rehabilitation monitoring. These systems allow joint angles, movement symmetry, and range of motion to be estimated using only a standard camera.

How Computer Vision Works for Estimation

RGB-camera-based computer vision systems use pose estimation models such as YOLO Pose or MediaPipe to detect human body keypoints directly from standard camera images. In this study, lower-body keypoints, including the hip, knee, and ankle, are extracted in real time to calculate joint angles, range of motion, movement trajectories, and symmetry-related measurements. Since RGB systems only require a normal camera, they are lower-cost, portable, and easier to deploy in rehabilitation environments compared with traditional motion capture systems.

Some existing studies have also explored the use of depth cameras, such as Microsoft Kinect, for lower-limb asymmetry analysis. Unlike RGB-only systems, depth cameras provide additional spatial distance information, allowing the system to estimate body movement in 3D space instead of relying only on 2D image coordinates. This can improve movement understanding and reduce some pose estimation ambiguities caused by occlusion or overlapping limbs. However, depth-camera systems require specialized hardware and are generally less portable than RGB-camera-based approaches. Since the final project focuses on accessibility and lower hardware cost, this study mainly investigates RGB-camera-based solutions while using depth-camera research as a comparison.

However, providing physiotherapists with a direct, non-technical, and visualized conclusion can make the system easier to use and can also give users immediate feedback. Therefore, an index called the Asymmetry Index will be used for this purpose.

Asymmetry Index

The Asymmetry Index, or ASI, is commonly used to quantify differences between the left and right sides of the body during movement. It directly represents imbalance as a percentage, allowing physiotherapists to evaluate the asymmetry condition of their patients more easily. In rehabilitation, ASI can be applied to measurements such as knee range of motion, repetition count, force output, gait timing, or estimated leg length.

A lower ASI value generally indicates better symmetry, while a higher ASI value may suggest imbalance, compensation, or incomplete recovery after injury.

Conclusion and How It Supports Our Project

This study suggests that lower-limb asymmetry can be estimated by combining computer vision data with ASI-based evaluation methods. Instead of only displaying raw joint angles or movement trajectories, ASI allows the system to summarize differences between the left and right lower limbs into a more direct and interpretable result for physiotherapists.

Although some existing solutions use depth cameras and advanced motion capture systems for higher accuracy, our Year 2 final project only uses an RGB camera due to portability and cost limitations. Therefore, depth-camera-based identification is not included in our system.

Overall, this independent study was successfully applied to the Year 2 project by displaying the ASI percentage on the website user interface. However, due to hardware limitations, we use an RGB camera with a YOLO model to detect joint positions and collect real-time data, which are then used to calculate reliable asymmetry results.