Systems and methods for detecting vehicle collisions
Systems, methods, models, and training for collision detection and near miss detection are discussed. Models are trained to identify collisions and near misses between two vehicles, based on how quickly a bounding box size changes for a vehicle represented in image data. Rapidly changing bounding box size is indicative of extreme proximity between vehicles, and thus indicative of collision or near-miss scenarios. Identifications of collisions or near misses can be reported to a remote operator, to send emergency assistance, perform driver evaluation, or alert other drivers in the area of risks.
1 . A method for identifying a collision between a first vehicle and second vehicle comprising:
accessing image data, the image data including a set of images, each image of the set of images captured by an image capture device positioned at the second vehicle;
applying, by at least one processor, an object detection model to each image in the set of images to determine a respective bounding box representation of the first vehicle from the perspective of the second vehicle for each image;
determining, by the at least one processor, for each image of the set of images, that the first vehicle and the second vehicle are positioned in a respective common lane of travel;
in response to determining that the first vehicle and the second vehicle are positioned in the respective common lane of travel:
determining, by the at least one processor, a rate of change of at least one spatial parameter of the bounding box for the first vehicle across at least a subset of the set of images, wherein the at least one spatial parameter is indicative of a size of the bounding box;
applying, by the at least one processor, a collision detection model to identify a collision between the first vehicle and the second vehicle based on the rate of change of the at least one spatial parameter of the bounding box; and
outputting an indication of collision when the collision detection model identifies a collision between the first vehicle and the second vehicle.
2 . The method of claim 1 , wherein if applying the object detection model to each image in the set of images to determine a respective bounding box representation of the first vehicle from the perspective of the second vehicle for each image does not result in a determination of any bounding boxes: identify no collision between the first vehicle and the second vehicle.
3 . The method of claim 1 , wherein accessing the image data comprises capturing the image data by the at least one image capture device positioned at the second vehicle.
4 . The method of claim 1 , wherein accessing the image data comprises retrieving the image data stored at a non-transitory processor-readable storage medium.
5 . The method of claim 1 , wherein the at least one processor is positioned at the second vehicle, and outputting an indication of collision comprises transmitting the indication of collision by a communication interface to be received by a device remote from the second vehicle.
6 . The method of claim 5 , further comprising outputting an alert to an operator of the device remote from the second vehicle.
7 . The method of claim 5 , further comprising transmitting, by a communication interface of the device remote from the second vehicle, the indication of collision to at least one other vehicle in a geographic region proximate a position of the second vehicle.
8 . A system for identifying a collision between a first vehicle and second vehicle comprising:
at least one processor;
at least one non-transitory processor-readable storage medium storing processor-executable instructions which when executed cause the system to:
access image data, the image data including a set of images, each image of the set of images captured by an image capture device positioned at the second vehicle;
apply, by the at least one processor, an object detection model to each image in the set of images to determine a respective bounding box representation of the first vehicle from the perspective of the second vehicle for each image;
determine, by the at least one processor, for each image of the set of images, whether the first vehicle and the second vehicle are positioned in a respective common lane of travel;
when the first vehicle and the second vehicle are positioned in the respective common lane of travel:
determine, by the at least one processor, a rate of change of at least one spatial parameter of the bounding box for the first vehicle across at least a subset of the set of images, wherein the at least one spatial parameter is indicative of a size of the bounding box;
apply, by the at least one processor, a collision detection model to identify a collision between the first vehicle and the second vehicle based on the rate of change of the at least one spatial parameter of the bounding box; and
output an indication of collision when the collision detection model identifies a collision between the first vehicle and the second vehicle; and
when the first vehicle and the second vehicle are not positioned in the respective common lane of travel:
determine, by the at least one processor that there is no collision between the first vehicle and the second vehicle.
9 . The system of claim 8 , wherein if the processor-executable instructions which cause the system to apply the object detection model to each image in the set of images to determine a respective bounding box representation of the first vehicle from the perspective of the second vehicle for each image does not result in a determination of any bounding boxes: the processor-executable instructions cause the system to identify no collision between the first vehicle and the second vehicle.
10 . The system of claim 8 , further comprising the at least one image capture device positioned at the second vehicle, wherein the processor-executable instructions which cause the system to access the image data cause the system to capture the image data by the at least one image capture device positioned at the second vehicle.
11 . The system of claim 10 , wherein the system includes a telematic device positioned at the second vehicle, and the at least one processor and the at least one non-transitory processor-readable storage medium are included in the telematic device.
12 . The system of claim 8 , wherein the processor executable instructions which cause the system to access the image data cause the system to retrieve the image data stored at the non-transitory processor-readable storage medium.
13 . The system of claim 8 , further comprising a communication interface at the second vehicle, wherein the at least one processor is positioned at the second vehicle, and the processor-executable instructions which cause the system to output an indication of collision cause the communication interface at the second vehicle to transmit the indication of collision to be received by a device remote from the second vehicle.
14 . The system of claim 13 , further comprising the device remote from the second vehicle, wherein the processor-executable instructions further cause the system to output, by an output interface of the device remote from the second vehicle, an alert to an operator of the device remote from the second vehicle.
15 . The system of claim 13 further comprising the device remote from the second vehicle, wherein the processor-executable instructions further cause the system to transmit, by a communication interface of the device remote from the second vehicle, the indication of collision to at least one other vehicle in a geographic region proximate a position of the second vehicle.
16 . The method of claim 1 , wherein determining that the first vehicle and the second vehicle are positioned in a respective common lane of travel comprises:
applying a feature detection model to identify road lanes;
identifying a lane of travel of the of the second vehicle; and
determining that the first vehicle is in the lane of travel of the second vehicle.
17 . The method of claim 1 , wherein determining that the first vehicle and the second vehicle are positioned in a respective common lane of travel comprises:
determining a left-distance from a left image edge and a right-distance from a right image edge for a representation of the first vehicle in the image of the set of images; and
determining that the first vehicle and the second vehicle are in a common lane of travel in response to a difference between the left-distance and the right-distance being within a horizontal distance threshold.
18 . The system of claim 8 , wherein the processor executable instructions which cause the at least one processor to determine whether the first vehicle and the second vehicle are positioned in a respective common lane of travel cause the at least one processor to:
apply a feature detection model to identify road lanes;
identify a lane of travel of the of the second vehicle; and
determine whether the first vehicle is in the lane of travel of the second vehicle.
19 . The system of claim 8 , wherein the processor-executable instructions which cause the at least one processor to determine whether the first vehicle and the second vehicle are positioned in a respective common lane of travel cause the at least one processor to:
determine a left-distance from a left image edge and a right-distance from a right image edge for a representation of the first vehicle in the image of the set of images; and
determine that the first vehicle and the second vehicle are in a common lane of travel if a difference between the left-distance and the right-distance are within a horizontal distance threshold.
20 . The method of claim 1 , wherein the collision detection model determines a confidence score indicative of a likelihood that a collision has occurred between the first vehicle and the second vehicle, and wherein outputting the indication of collision comprises outputting the indication of collision when the collision detection model determines the confidence score which satisfies a confidence threshold.
21 . The system of claim 8 , wherein the collision detection model determines a confidence score indicative of a likelihood that a collision has occurred between the first vehicle and the second vehicle, and wherein the processor-executable instructions which cause the system to output the indication of collision cause the system to output the indication of collision when the collision detection model determines the confidence score which satisfies a confidence threshold.