Systems and methods for estimating future paths

Patent metadata

Field	Value
Publication number	US-2017193338-A1
Application number	US-201715398926-A
Country	US
Kind code	A1
Filing date	Jan 5, 2017
Priority date	Jan 5, 2016
Publication date	Jul 6, 2017
Grant date	—

Abstract

A system and method estimate a future path ahead of a current location of a vehicle. The system includes at least one processor programmed to: obtain an image of an environment ahead of a current arbitrary location of a vehicle navigating a road; obtain a trained system that was trained to estimate a future path on a first plurality of images of environments ahead of vehicles navigating roads; apply the trained system to the image of the environment ahead of the current arbitrary location of the vehicle; and provide, based on the application of the trained system to the image, an estimated future path of the vehicle ahead of the current arbitrary location.

First claim

What is claimed is:

1 . A system for estimating a future path ahead of a current location of a vehicle, comprising:
at least one processor programmed to: obtain an image of an environment ahead of a current arbitrary location of a vehicle navigating a road; obtain a trained system that was trained to estimate a future path on a first plurality of images of environments ahead of vehicles navigating roads; apply the trained system to the image of the environment ahead of the current arbitrary location of the vehicle; and provide, based on application of the trained system to the image, an estimated future path of the vehicle ahead of the current arbitrary location.

2 . The system according to claim 1 , wherein the trained system comprises piece-wise affine functions of global functions.

3 . The system according to claim 2 , wherein the global functions comprise convolutions, max pooling, or a rectifier liner unit.

4 . The system according to claim 2 , wherein the at least one processor is further programmed to: utilize the estimated future path ahead of the current location of the vehicle to control at least one electronic or mechanical unit of the vehicle to change at least one motion parameter of the vehicle.

5 . The system according to claim 2 , wherein the at least one processor is further programmed to: utilize the estimated future path ahead of the current location of the vehicle to provide a sensory feedback to a driver of the vehicle.

6 . The system according to claim 1 , wherein the estimated future path of the vehicle ahead of the current location is further based on identifying one or more predefined objects appearing in the image of the environment using at least one classifier.

7 . The system according to claim 1 , wherein the at least one processor is further programmed to: utilize the estimated future path ahead of the current location of the vehicle to provide a control point for a steering control function of the vehicle.

8 . The system according to claim 1 , wherein applying the trained system to the image of the environment ahead of the current location of the vehicle provides two or more estimated future paths of the vehicle ahead of the current location.

9 . The system according to claim 1 , wherein the at least one processor is further programmed to: utilize the estimated future path ahead of the current location of the vehicle in estimating a road profile ahead of the current location of the vehicle.

10 . The system according to claim 1 , wherein applying the trained system to the image of the environment ahead of the current location of the vehicle provides two or more estimated future paths of the vehicle ahead of the current location, and further comprising estimating a road profile along each one of the two or more estimated future paths of the vehicle ahead of the current location.

11 . The system according to claim 1 , wherein the at least one processor is further programmed to: utilize the estimated future path ahead of the current location of the vehicle in detecting one or more vehicles located in or near the future path of the vehicle.

12 . The system according to claim 11 , wherein the at least one processor is further programmed to: cause at least one electronic or mechanical unit of the vehicle to change at least one motion parameter of the vehicle based on a location of one or more vehicles which were determined to be in or near the future path of the vehicle.

13 . The system according to claim 11 , wherein the at least one processor is further programmed to: trigger a sensory alert to indicate to a user of that one or more vehicles are determined to be in or near the future path of the vehicle.

14 . A method of processing images, comprising:
obtaining a first plurality of training images, wherein each one of the first plurality of training images is an image of an environment ahead of a vehicle navigating a road; for each one of the first plurality of training images, obtaining a prestored path of the vehicle ahead of a respective present location of the vehicle; training a system to provide, given an image, a future path for a vehicle navigating a road ahead of a respective present location of the vehicle, wherein training the system comprises:
providing the first plurality of training images as input to the trained system;
at each iteration of the training, computing a loss function based on a respective provisional future path that was estimated by a current state of weights of the trained system and a respective prestored path; and
updating the weights of the trained system according to results of the loss function.

15 . The method according to claim 14 , wherein obtaining the first plurality of training images further comprises, obtaining, for each one of the images from the first plurality of training images, data indicating a location of the vehicle on the road at an instant when the image was captured.

16 . The method according to claim 15 , wherein obtaining the first plurality of training images, comprises obtaining a location of at least one lane mark in at least one image from the first plurality of training images, and wherein obtaining, for each one of the images from first plurality of training images, data indicating the location of the vehicle on the road at the instant when the image was captured, comprises, for the at least one image from the first plurality of training images, determining the location of the vehicle on the road at an instant when the at least one image was captured according to a location of the at least one lane mark in the at least one image.

17 . The method according to claim 16 , wherein determining the location of the vehicle on the road at an instant when the at least one image from the first plurality of training images was captured according to a location of the at least one lane mark in the at least one image, comprises determining the location of the vehicle on the road at a predefined offset from the location of the at least one lane mark.

18 . The method according to claim 15 , wherein the prestored path of the vehicle ahead of the respective present location of the vehicle is determined based on locations of the vehicle on the road at respective instants when a respective second plurality of training images were captured, and wherein the second plurality of training images are images from the first plurality of training images that were captured subsequent to the image associated with the present location.

19 . The method according to claim 14 , wherein training the system includes a plurality of iterations and is carried out until a stop condition is met.

20 . The method according to claim 14 , further comprising, providing as output a trained system that is configured to provide, given an arbitrary input image of an environment ahead of a vehicle navigating a road, a future path estimation for the vehicle.

21 . The method according to claim 14 , wherein the first plurality of training images includes a relatively higher number of images of environments which appear relatively rarely on roads.

22 . The method according to claim 21 , wherein the first plurality of training images includes a relatively higher number of images of environments that comprise a curved road.

23 . The method according to claim 21 , wherein the first plurality of training images includes a relatively higher number of images of environments that comprise

Assignees

• Mobileye Vision Technologies Ltd

Inventors

• Huberman David
• SHAMMAH Shaked

Classifications

• G06V10/82 Primary

  using neural networks · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/24317

  Piecewise classification, i.e. whereby each classification requires several discriminant rules · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N20/00

  Machine learning · CPC title