System and methods for detecting abnormal following vehicles
Embodiments of the present disclosure a privacy-preserving defensive driving system that can detect abnormal following vehicles during driving. An example system may be configured to: continuously capture video data of the camera's field-of-view, detect following vehicles in the captured video data, and determine whether one or more following vehicles is exhibiting abnormal following behavior with respect to a first vehicle.
1 . A system for detecting abnormal following vehicles, the system comprising:
at least one sensor configured to detect motion data of a first vehicle along the x-axis, y-axis, and z-axis;
a camera positioned within the first vehicle such that a field-of-view (FOV) of the camera faces outward from a rear of the first vehicle;
a user interface;
a processor;
memory having instructions stored thereon that, when executed by the processor, cause the system to:
continuously capture video data of the camera's FOV;
detect following vehicles in the captured video data using an object detection model;
determine an amount of time that each of the following vehicles follows the first vehicle, wherein the amount of time (T ID ) that each of the following vehicles follows the first vehicle is calculated as:
T
ID
=
N
L
-
N
F
f
r
where f r denotes a frame rate of the camera, N F denotes a time that the respective following vehicle was first detected in the video data, and N L denotes a time that the respective following vehicle is no longer detected in the video data;
determine critical driving behavior of the first vehicle based at least in part on a measure of variation in the detected motion data over the y-axis or z-axis, wherein determining the critical driving behavior comprises filtering noise and removing road condition artifacts from the motion data across at least one coordinate system axis;
perform a sensor fusion operation to synchronize the video data and the filtered motion data based on detected time variations to determine whether one or more of the following vehicles is exhibiting abnormal following behavior with respect to the first vehicle based on the amount of time that each of the following vehicles follows the first vehicle and the determined critical driving behavior of the first vehicle; and
responsive to determining that one or more of the following vehicles is exhibiting abnormal following behavior, generate and output an audio or visual alert or output driving instructions to a designated safe location via the user interface.
2 . The system of claim 1 , wherein the at least one sensor comprises an inertial measurement unit (IMU).
3 . The system of claim 1 , wherein the at least one sensor, camera, processor, user interface, and memory are components of a smartphone.
4 . The system of claim 1 , wherein one or more of the at least one sensor, camera, processor, user interface, and memory are components of a vehicle computer.
5 . The system of claim 1 , wherein the object detection model is a deep convolution neural network.
6 . The system of claim 1 , wherein the object detection model is a You Only Look Once (YOLO) algorithm.
7 . The system of claim 1 , wherein the noise is filtered from the motion data using a Savitzky-Golay filter.
8 . The system of claim 1 , wherein to determine whether one or more of the following vehicles is exhibiting abnormal following behavior with respect to the first vehicle based on the amount of time that each of the following vehicles follows the first vehicle and the critical driving behavior of the first vehicle, the instructions cause the system to:
determine an anomaly score for each of the following vehicles based on the critical driving behavior of the first vehicle within respective amounts of time that each of the following vehicles were following the first vehicle.
9 . The system of claim 8 , wherein the anomaly scores are determined using a Local Outlier Factor (LOF) algorithm.
10 . The system of claim 1 , wherein the instructions further cause the system to alert an operator of the first vehicle if one or more of the following vehicles is determined to exhibit abnormal following behavior.
11 . A method for detecting abnormal following vehicles, the method comprising:
obtaining motion data for a first vehicle via at least one sensor along the x-axis, y-axis, and z-axis;
continuously capturing video data of a rear FOV of the first vehicle via a second sensor;
detecting one or more following vehicles in the captured video data using an object detection model;
determining an amount of time that each of the following vehicles follows the first vehicle, wherein the amount of time (T ID ) that each of the following vehicles follows the first vehicle is calculated as:
T
ID
=
N
L
-
N
F
f
r
where f r denotes a frame rate of the second sensor, N F denotes a time that the respective following vehicle was first detected in the video data, and N L denotes a time that the respective following vehicle is no longer detected in the video data;
determining critical driving behavior of the first vehicle based at least in part on a measure of variation in the motion data over the y-axis or z-axis, wherein determining the critical driving behavior comprises filtering noise and removing road condition artifacts from the motion data across at least one coordinate system axis;
performing a sensor fusion operation to synchronize the video data and the filtered motion data based on detected time variations to determine whether one or more of the following vehicles is exhibiting abnormal following behavior with respect to the first vehicle based on the amount of time that each of the following vehicles follows the first vehicle and the determined critical driving behavior of the first vehicle; and
responsive to determining that one or more of the following vehicles is exhibiting abnormal following behavior, generating and outputting an audio or visual alert or outputting driving instructions to a designated safe location via a user interface.
12 . The method of claim 11 , wherein the object detection model comprises at least one of a deep convolution neural network or a YOLO algorithm.