Inertial-Sensor Video Tracking for Overlapping Athletes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis tools struggle to reliably track multiple overlapping athletes in video footage due to obscuration, collisions, and lack of distinguishing features, leading to inconsistent tracking.
Innovation Solution
A system that combines machine learning models with motion data from wearable monitors to identify and assign persistent identifiers to athletes, correlating video events with motion events to enhance tracking accuracy and correct time discrepancies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If image analysis tools are used to track athletes in video footage, then tracking can be performed without additional sensors, but tracking reliability deteriorates when athletes are obscured or collide
Solution Approach 1:
The patent combines video footage analysis with data from wearable motion sensors to create a hybrid tracking system. The video provides visual context while the sensors provide reliable motion data even when athletes are obscured, merging the strengths of both approaches to overcome their individual limitations.
Solution Approach 2:
The wearable motion sensors act as intermediaries that provide alternative means of tracking when visual tracking fails. The system uses sensor data to supplement and verify video-based tracking, particularly during occlusions or collisions where visual information is insufficient.
2Productivity
If multiple athletes are tracked simultaneously in the same frame, then comprehensive team performance analysis is enabled, but tracking precision deteriorates due to overlapping and obscuration
Solution Approach 1:
The system segments the tracking problem by assigning individual motion sensor data streams to each athlete, allowing precise identification even when visual appearances overlap. Each sensor provides unique motion signatures that distinguish individual athletes regardless of their positions or occlusions in the video frame.
Solution Approach 2:
The system adds a temporal dimension to tracking by analyzing motion patterns over time. Rather than relying solely on spatial separation in video frames, the system uses time-series motion data from sensors to distinguish and track athletes through their unique movement patterns, even when they occupy the same spatial location.
3Device complexity
If traditional video tracking is used, then no additional hardware is required, but tracking consistency deteriorates during occlusions and collisions
Solution Approach 1:
The system performs preliminary actions by having athletes wear motion sensors before the sporting event begins. This ensures that motion data is continuously collected and available from the start, allowing the system to maintain tracking consistency during occlusions and collisions without needing to reactively add hardware during the event.
Data Source
Figure 1A
Figure 1B
Figure 1C~1D
AI summary
A method of tracking video objects within video footage includes receiving, at a computing device, video footage of a space and event data from a plurality of monitors. Each monitor includes a motion sensor configured to measure movements of a respective monitored video object within the space. The event data includes motion events transmitted by the monitors. The method also includes using a machine learning model to identify and track the monitored video objects within the video footage and to identify, from the video footage, video events performed by the monitored video objects. The method also includes using the computing device to assign a respective persistent identifier to each of the monitored video objects within the video footage based at least in part on commonality between video events performed by the monitored video object and motion events transmitted by one of the monitors.