First Approach to Automatic Measurement of Frontal Plane Projection Angle During Single Leg Landing Based on Depth Video

First Approach to Automatic Measurement of Frontal Plane Projection Angle During Single Leg Landing Based on Depth VideoUCAmI 2016 (Las Palmas de Gran Canaria, Spain)

Carlos Bailon1, Miguel Damas1, Hector Pomares1 and Oresti Banos2

1Department of Computer Architecture and Computer Technology, CITIC-UGR Research Center, University of Granada, Spain

2Telemedicine Cluster of the Biomedical Signal and Systems Group, University of Twente, Netherlands

Knee alignment• Grade of alignment of the hip, knee and ankle joints.• Commonly used as a risk indicator of many biomechanical

injuries related to knee joint when measured during the performance dynamic tasks.

Anterior Cruciate Ligament (ACL) injuriesPatellofemoral Pain Syndrome (PFPS)

Potential misalignments during dynamic exercises are the most common injury mechanisms.

Quantification of knee alignmentProjection of the angle formed by the hip, knee and ankle joints over the frontal plane of the body.

Frontal Plane Projection Angle

(FPPA)Wilson et al. “Core strength and lower extremity alignment during single leg squats” Medicine & Science in Sports & Exercise (2006)

Key limitations of existing techniques for FPPA measuring

Inertial sensor-based systems

3D motion tracking video systems

2D offline video analysis

Accurate 3D rotationsPossible motion restrictionNon-deliberated sensor displacement

Tridimensional motion tracking

High sampling rateNeed of high number of camerasCostly and space demanding

One camera needed

Portable and easy-to-use equipmentElevated time for analysisProne to human errors2D analysis

Objectives of the project• Automatic estimation of FPPA during the

performance of dynamic tasks (ideally any 2D biomechanics angle)

• Single-camera solution.

• No external light sources.

• Inexpensive and easy-to-use system.

• Real-time visualization of the FPPA.

• Automatic analysis of the data.

Overview of the proposed system

Why do we use markers?Although Kinect is well-known for being a markerless system, we introduce the tracking of three retro-reflective markers.

This method increases the accuracy of the pose estimation algorithm of Kinect and allows for tracking points that are not necessarily joints.The blue line shows the data registered during a single leg landing using markers.

The red line shows the data registered using the Kinect pose estimation algorithm.

RMSE = 8.498º

Reflective markers tracking• Kinect’s depth sensor

captures the infrared intensity value for each pixel of the image (512 x 424 resolution).

• An empirical intensity threshold (high-pass filter) is used to select candidate marker’s pixels.

• Kinect’s pose estimation algorithm is used to classify each marker position.

• Retro-reflective elements not belonging to a marker are ignored.

• Markers coordinates are computed as the midpoints of the selected pixel’s cloud.

Application

Application

Experimental results

High concordance among the measurements Proposed approach saves up to 10 minutes per

assessed subject

Comparison between Kinovea (2D offline analysis tool, expert oriented) and the proposed system. FPPA evaluated for 10 healthy subjects from a professional football team

Conclusions• Proposed a novel system to perform an

automatic estimation of dynamic FPPA, by a single-camera, cost-effective and portable solution.

• The system uses a depth sensor to track the position of three retro-reflective markers attached to the subject’s hip, knee and ankle joints.

• Designed a user interface which simplifies the expert’s routine and expedites the analysis of the results.

• Experimental results show the interrater reliability of the proposed system, as well as the limitations of the 2D analysis (limited joint rotation measurement).

THANKS!

Application description

Application implementation

Data storageLocal database engineWhy?

• On-disk database file.• Not very large dataset.• No concurrent writers.• Data easily exported to CSV files for external

analysis.

Data is stored in two tables, differentiating patient personal information and data collected. Both tables are related by a personal ID.