Crash Detection on Mobile Devices Using Multimodal Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In severe automobile accidents, drivers or passengers may be incapacitated, preventing them from using their mobile devices to call for help, especially in isolated locations without bystanders.
Innovation Solution
A method for crash detection on mobile devices, such as smartwatches and smartphones, that uses multimodal features from various sensors to detect crashes, determines the severity of the crash, and automatically contacts emergency services if the user does not respond within a predetermined time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If automatic crash detection is implemented using multiple sensors and machine learning models, then the reliability of crash detection is improved, but the device complexity increases
Solution Approach 1:
The crash detection system is divided into multiple independent sensing modalities (accelerometer, gyroscope, barometer, microphone) that each detect specific crash characteristics. This segmentation allows the system to improve detection reliability through multiple independent verification channels while managing complexity by keeping each sensor's processing relatively simple and modular.
Solution Approach 2:
Multiple sensing modalities and machine learning models are merged into a unified crash detection system. The accelerometer, gyroscope, barometer, and microphone data are combined and processed together through machine learning models to make a comprehensive crash determination, improving reliability by leveraging complementary information from different sensors.
2Measurement precision
If multiple sensing modalities and machine learning models are used to reduce false alarms, then the measurement precision of crash detection is improved, but the use of energy increases
Solution Approach 1:
The system applies machine learning models selectively based on detected conditions. When a potential crash is detected through initial sensor thresholds, then the full machine learning analysis is activated. During normal operation, the system uses simpler threshold-based detection, reducing energy consumption while maintaining measurement precision when it matters most.
Solution Approach 2:
Traditional mechanical threshold-based crash detection is replaced with machine learning models that analyze sensor data patterns. This substitution improves measurement precision by recognizing complex crash signatures that simple thresholds would miss, while the system manages energy by using the computationally intensive models only when needed.
3Reliability
If the system waits for user response before contacting emergency services, then false alarms are reduced, but the loss of time in severe crashes increases
Solution Approach 1:
The system presents a preliminary alert to the user immediately upon detecting a crash, giving them a chance to dismiss false alarms before emergency services are contacted. This preliminary user confirmation step reduces false alarms while minimizing time loss by having the response mechanism ready and waiting, rather than delaying the entire emergency response.
Solution Approach 2:
The system implements a feedback loop where the user's response to the crash alert is used to determine whether to proceed with emergency service contact. This feedback mechanism allows the system to distinguish between true crashes and false alarms by checking user intent, balancing reliability with rapid response through automated escalation if no response is received.
Data Source
AI summary
Embodiments are disclosed for crash detection on one or more mobile devices (e.g., smartwatch and/or smartphone. In some embodiments, a method comprises: detecting a crash event on a crash device; extracting multimodal features from sensor data generated by multiple sensing modalities of the crash device; computing a plurality of crash decisions based on a plurality of machine learning models applied to the multimodal features, wherein at least one multimodal feature is a rotation rate about a mean axis of rotation; and determining that a severe vehicle crash has occurred involving the crash device based on the plurality of crash decisions and a severity model.