Lane recognition apparatus

• US 7,889,887 B2
• Filed: 03/16/2006
• Issued: 02/15/2011
• Est. Priority Date: 03/18/2005
• Status: Expired due to Fees

• Current Assignee: Honda Elesys Co., Ltd. (Nidec Corporation)
• Original Assignee: Honda Elesys Co., Ltd. (Nidec Corporation)
• Inventors: Azuma, Takahiro
• Primary Examiner(s): Le; Vu
• Assistant Examiner(s): Wang; Claire
• Application Number:
  US11/376,101
• Publication Number:
  US 20060210116A1
• Time in Patent Office: 1,797 Days
• Field of Search: 382/104, 382/274, 382/170, 382/106, 348/148, 348/143
• US Class Current: 382/104
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
• CPC Class Codes: G06V 20/588 Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road

Abstract

A lane recognition apparatus that recognizes a lane from an input image sequence captured from a visual sensor mounted on a vehicle at a fixed time interval, includes a time series smoothing part for generating a smoothed feature image based on a feature image extracted from the input image sequence so that a dotted line lane boundary describes a linear locus over time, a gradient image generation part for generating a gradient image by estimating a feature value gradient at each pixel position of the smoothed feature image, and a Hough transform part for performing Hough transform on the gradient image and detecting lines of edges in the smoothed feature image.

Claims

1. A lane recognition apparatus that recognizes a dashed or dotted line lane boundary from an input image sequence captured from a visual sensor mounted on a vehicle at a fixed time interval, comprising:

a time series smoothing part configured to generate a smoothed feature image based on a feature image extracted from the input image sequence such that a dashed or dotted line lane boundary describes a linear locus over time;

a gradient image generation part configured to generate a gradient image by estimating a feature value gradient at each pixel position of the smoothed feature image;

a Hough transform part configured to perform a Hough transform on the gradient image and to detect lines of edges in the smoothed feature image, wherein the time series smoothing part, the gradient image generation part, and the Hough transform part are implemented by at least one of a large scale integrated circuit or a CPU; and

a feature image extraction part configured to extract the feature image indicating a likelihood of the lane boundary by a feature value based on peripheral pixels at each pixel position of a projection image of a road surface and to supply the feature image to the time series smoothing part,

wherein the feature image extraction part takes pixel values of periphery of each position of a road surface projection image as an input to an image filter, which outputs a real number value or an integral number value indicating positions where a likelihood of existence of a lane marker such as a white line is high.

2. A lane recognition apparatus according to claim 1, further comprising:

an inverse perspective projection part configured to project each image of the input image sequence on a plane coordinate system that approximates the road surface in accordance with an optical model of a sensor and to supply the projection image of the road surface to the time series smoothing part.

3. A lane recognition apparatus according to claim 1, wherein the time series smoothing part is configured to add the feature image at a certain time into the smoothed feature image at the preceding time at a predetermined rate to update the smoothed feature image to obtain a smoothed feature image smoothed in time series.

4. A lane recognition apparatus according to claim 3, wherein the time series smoothing part is further configured to add a predetermined arithmetic formula having a spatial frequency response to the smoothed feature image in order to emphasize a dashed or dotted line lane boundary that describes a linear locus over time.

9. A lane recognition apparatus according to claim 1, wherein the time series smoothing part is configured to perform time series smoothing such that a signal of a linear pattern largely deviating from a traveling direction is weakened.

5. A lane recognition apparatus that recognizes a dashed or dotted line lane boundary from an input image sequence captured from a visual sensor mounted on a vehicle at a fixed time interval, comprising:

time series smoothing means for generating a smoothed feature image based on a feature image extracted from the input image sequence such that a dashed or dotted line lane boundary describes a linear locus over time;

gradient image generation means for generating a gradient image by estimating a feature value gradient at each pixel position of the smoothed feature image; and

Hough transform means for performing a Hough transform on the gradient image and to detect lines of edges in the smoothed feature image; and

feature image extraction means for extracting the feature image indicating a likelihood of the lane boundary by a feature value based on peripheral pixels at each pixel position of a projection image of a road surface and to supply the feature image to the time series smoothing means,

wherein the feature image extraction means takes pixel values of periphery of each position of a road surface projection image as an input to an image filter, which outputs a real number value or an integral number value indicating positions where a likelihood of existence of a lane marker such as a white line is high.

6. A lane recognition apparatus according to claim 5, further comprising:

inverse perspective projection means for projecting each image of the input image sequence on a plane coordinate system that approximates the road surface in accordance with an optical model of a sensor and to supply the projection image of the road surface to the time series smoothing means.

7. A lane recognition apparatus according to claim 5, wherein the time series smoothing means is configured to add the feature image at a certain time into the smoothed feature image at the preceding time at a predetermined rate to update the smoothed feature image to obtain a smoothed feature image smoothed in time series.

8. A lane recognition apparatus according to claim 7, wherein the time series smoothing means is further configured to add a predetermined arithmetic formula having a spatial frequency response to the smoothed feature image in order to emphasize a dashed or dotted line lane boundary that describes a linear locus over time.

10. A lane recognition apparatus according to claim 5, wherein the time series smoothing means is configured to perform time series smoothing such that a signal of a linear pattern largely deviating from a traveling direction is weakened.