Automated assessment of collision risk based on computer vision
An image may be obtained from one or more cameras coupled to a first vehicle. The image may be provided as input to a machine learning algorithm configured to determine whether an object depicted in the image corresponds to another vehicle and to determine size information and location information for the object. Output from the machine learning algorithm enables obtaining features including size and location information for a second vehicle that is identified in the image. The features may be used to determine whether the second vehicle is depicted within a predetermined region of the image including a predicted travel path of the first vehicle. The features may also be used to determine whether the second vehicle is within a predetermined proximity of the first vehicle. Thereafter, a determination may be generated as to whether there is a significant risk of collision between the first vehicle and the second vehicle.
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing a plurality of instructions which, when executed by the one or more processors, cause the system to perform steps comprising:
obtaining an image from one or more cameras coupled to a first vehicle, wherein obtaining the image is responsive to one or more non-camera sensors detecting an object outside the first vehicle;
providing the image as input to a first machine learning algorithm, wherein the first machine learning algorithm is configured to determine whether or not the object depicted in the image corresponds to another vehicle and to determine size information and location information for the object;
obtaining a set of features based on output from the first machine learning algorithm, wherein the set of features includes size information and location information for a second vehicle that is identified in the image;
determining, based on the set of features, whether or not the second vehicle is depicted within a predetermined region of the image, wherein the predetermined region includes a predicted travel path of the first vehicle;
determining, based on the set of features, whether or not the second vehicle is within a predetermined proximity of the first vehicle; and
generating, based on determining whether or not the second vehicle is depicted within the predetermined region and based on determining whether or not the second vehicle is within the predetermined proximity, a first determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle.
2. The system of claim 1 , wherein the plurality of instructions further comprise instructions which, when executed by the one or more processors, cause the system to perform steps comprising: after obtaining the set of features based on the output from the first machine learning algorithm, providing the set of features as input to a second machine learning algorithm that is configured to generate a second determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle; obtaining the second determination as output from the second machine learning algorithm; and performing an ensembling process on the first determination and the second determination to determine whether or not there is a significant risk of collision between the first vehicle and the second vehicle.
3. The system of claim 1 , wherein the first machine learning algorithm is performed using a convolutional neural network.
4. The system of claim 1 , wherein the predetermined region has a triangular shape.
5. The system of claim 1 , wherein the size information comprises one or more dimension values of a bounding box that includes the second vehicle.
6. The system of claim 1 , wherein the location information comprises a pair of coordinates corresponding to an approximate center of a bounding box that includes the second vehicle.
7. A method comprising:
obtaining an image from one or more cameras coupled to a first vehicle, wherein obtaining the image is responsive to one or more non-camera sensors detecting an object outside the first vehicle; providing the image as input to a first machine learning algorithm, wherein the first machine learning algorithm is configured to determine whether or not the object depicted in the image corresponds to another vehicle and to determine size information and location information for the object; obtaining a set of features based on output from the first machine learning algorithm, wherein the set of features includes size information and location information for a second vehicle that is identified in the image; determining, based on the set of features, whether or not the second vehicle is depicted within a predetermined region of the image, wherein the predetermined region includes a predicted travel path of the first vehicle; determining, based on the set of features, whether or not the second vehicle is within a predetermined proximity of the first vehicle; and generating, based on determining whether or not the second vehicle is depicted within the predetermined region and based on determining whether or not the second vehicle is within the predetermined proximity, a first determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle, wherein the method is performed by one or more computing devices.
8. The method of claim 7 , further comprising:
after obtaining the set of features based on the output from the first machine learning algorithm, providing the set of features as input to a second machine learning algorithm that is configured to generate a second determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle; obtaining the second determination as output from the second machine learning algorithm; and performing an ensembling process on the first determination and the second determination to determine whether or not there is a significant risk of collision between the first vehicle and the second vehicle.
9. The method of claim 7 , wherein the first machine learning algorithm is performed using a convolutional neural network.
10. The method of claim 7 , wherein the predetermined region has a triangular shape.
11. The method of claim 7 , wherein the size information comprises one or more dimension values of a bounding box that includes the second vehicle.
12. The method of claim 7 , wherein the location information comprises a pair of coordinates corresponding to an approximate center of a bounding box that includes the second vehicle.
13. One or more non-transitory computer-readable media storing a plurality of instructions configured for execution by one or more processors, the plurality of instructions comprising:
obtaining an image from one or more cameras coupled to a first vehicle, wherein obtaining the image is responsive to one or more non-camera sensors detecting an object outside the first vehicle;
providing the image as input to a first machine learning algorithm, wherein the first machine learning algorithm is configured to determine whether or not the object depicted in the image corresponds to another vehicle and to determine size information and location information for the object;
obtaining a set of features based on output from the first machine learning algorithm, wherein the set of features includes size information and location information for a second vehicle that is identified in the image;
determining, based on the set of features, whether or not the second vehicle is depicted within a predetermined region of the image, wherein the predetermined region includes a predicted travel path of the first vehicle;
determining, based on the set of features, whether or not the second vehicle is within a predetermined proximity of the first vehicle; and
generating, based on determining whether or not the second vehicle is depicted within the predetermined region and based on determining whether or not the second vehicle is within the predetermined proximity, a first determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle.
14. One or more non-transitory computer-readable media storing a plurality of instructions configured for execution by one or more processors, the plurality of instructions comprising:
providing an image as input to a first machine learning algorithm, wherein the image is obtained from one or more cameras coupled to a first vehicle, wherein the first machine learning algorithm is configured to determine whether or not an object depicted in the image corresponds to another vehicle and to determine size information and location information for the object;
obtaining a set of features based on output from the first machine learning algorithm, wherein the set of features includes size information and location information for a second vehicle that is identified in the image;
determining, based on the set of features, whether or not the second vehicle is depicted within a predetermined region of the image, wherein the predetermined region includes a predicted travel path of the first vehicle;
determining, based on the set of features, whether or not the second vehicle is within a predetermined proximity of the first vehicle;
generating, based on determining whether or not the second vehicle is depicted within the predetermined region and based on determining whether or not the second vehicle is depicted within the predetermined region and based on determining whether or not the second vehicle is within the predetermined proximity, a first determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle;
providing the set of features as input to a second machine learning algorithm that is configured to generate a second determination as to whether or not there is a significant risk of collision between the first vehicle and the second vehicle;
obtaining the second determination as output from the second machine learning algorithm; and
performing an ensembling process on the first determination and the second determination to determine whether or not there is a significant risk of collision between the first vehicle and the second vehicle.
15. The one or more non-transitory computer-readable media of claim 13 , wherein the first machine learning algorithm is performed using a convolutional neural network.
16. The one or more non-transitory computer-readable media of claim 13 , wherein the predetermined region has a triangular shape.
17. The one or more non-transitory computer-readable media of claim 13 , wherein the size information comprises one or more dimension values of a bounding box that includes the second vehicle.
18. The one or more non-transitory computer-readable media of claim 13 , wherein the location information comprises a pair of coordinates corresponding to an approximate center of a bounding box that includes the second vehicle.