Method and system for vehicular collision reconstruction
A system for accident reconstruction can include and/or be configured to interface with any or all of: a set of models, a set of modules, a processing system, client application, a user device (equivalently referred to herein as a mobile device), a set of sensors, a vehicle, and/or any other suitable components. A method for accident reconstruction includes collecting a set of inputs; detecting a collision and/or one or more features of the collision; reconstructing the collision; and producing an output based on the reconstruction. Additionally or alternatively, the method can include training a set of models and/or modules, and/or any other suitable processes.
1. A method for detecting and characterizing a collision based on a set of mobile device sensor data, the method comprising:
training a set of multiple models to produce a set of trained models;
with a Software Development Kit (SDK) operating at a mobile device, collecting the set of mobile device sensor data from a set of sensors onboard the mobile device;
evaluating the set of trained models, comprising:
checking for a detected collision based on a set of outputs of a first portion of the set of trained models;
producing a first confidence level associated with a detected collision;
in response to the first confidence level falling below a predetermined threshold, iteratively repeating evaluation of the first portion of the set of trained models to produce a set of additional confidence levels associated with the detected collision;
in response to at least one of the set of additional confidence levels exceeding the predetermined threshold, characterizing a set of features of the detected collision with a second portion of the set of trained models, the set of features comprising an identification of a collision type, the collision type comprising one of: a frontal impact, a rear impact, a side impact, or a rollover impact;
in response to detecting the collision and characterizing the set of features, retraining each of the set of trained models;
wherein the set of trained models comprises a set of gradient boosting machines.
2. The method of claim 1 , wherein the set of sensor data comprises Global Positioning System (GPS) data, barometer data, and motion data.
3. The method of claim 1 , wherein the set of trained models comprises a set of multiple architectures.
4. The method of claim 3 , wherein the set of multiple architectures comprises at least one of: a gradient boosting machine or a convolutional neural network.
5. The method of claim 1 , wherein the set of trained models comprises a set of deep learning models.
6. The method of claim 5 , wherein the set of deep learning models comprises a convolutional neural network.
7. The method of claim 1 , wherein characterizing the set of features further comprises producing a predicted likelihood of fraud metric associated with the detected collision.
8. The method of claim 7 , wherein the predicted likelihood of fraud metric is determined based on features of at least one of braking or acceleration as determined with at least one of the second portion of the set of trained models and based on the set of mobile device sensor data.
9. The method of claim 7 , wherein the method further comprises, in response to the predicted likelihood of fraud metric exceeding a predetermined threshold, triggering collection of additional data after a trip associated with the detected collision has ended.
10. A system for detecting and characterizing a collision based on a set of mobile device sensor data, the system comprising:
a set of trained models, the set of trained models comprising multiple trained models;
a Software Development Kit (SDK) operating at a mobile device, wherein the SDK:
collects the set of mobile device sensor data from a set of sensors onboard the mobile device;
a processor in communication with the SDK, wherein the processor:
trains the set of trained models;
evaluates the set of trained models, comprising:
checking for a detected collision based on a set of outputs of a first portion of the set of trained models;
producing a first confidence level associated with a detected collision;
In response to the first confidence level falling below a predetermined threshold, iteratively repeating evaluation of the first portion of the set of trained models to produce a set of additional confidence levels associated with the detected collision;
in response to at least one of the set of additional confidence levels exceeding the predetermined threshold, characterizing a set of features of the detected collision with a second portion of the set of trained models, the set of features comprising an identification of a collision type, the collision type comprising one of: a frontal impact, a rear impact, a side impact, or a rollover impact;
in response to detecting the collision and characterizing the set of features, retraining each of the set of trained models;
wherein the set of trained models comprises a set of gradient boosting machines.
11. The system of claim 10 , wherein the set of sensor data comprises Global Positioning System (GPS) data, barometer data, and motion data.
12. The system of claim 10 , wherein the set of trained models comprises a set of multiple architectures.
13. The system of claim 12 , wherein the set of multiple architectures comprises at least one of: a gradient boosting machine or a convolutional neural network.
14. The system of claim 10 , wherein the set of trained models comprises a set of deep learning models.
15. The system of claim 14 , wherein the set of deep learning models comprises a convolutional neural network.
16. The system of claim 10 , wherein characterizing the set of features further comprises producing a predicted likelihood of fraud metric associated with the detected collision.
17. The system of claim 16 , wherein the predicted likelihood of fraud metric is determined based on features of at least one of braking or acceleration as determined with at least one of the second portion of the set of trained models and based on the set of mobile device sensor data.
18. The system of claim 16 , wherein the processor further, in response to the predicted likelihood of fraud metric exceeding a predetermined threshold, triggers collection of additional data after a trip associated with the detected collision has ended.