Image-processing method and apparatus for recognizing objects in traffic

• US 6,556,692 B1
• Filed: 07/14/1999
• Issued: 04/29/2003
• Est. Priority Date: 07/14/1998
• Status: Expired due to Fees

• Current Assignee: Daimler AG (Mercedes-Benz Group AG)
• Original Assignee: DaimlerChrysler AG (Mercedes-Benz Group AG)
• Inventors: Gavrila, Dariu
• Primary Examiner(s): Mehta, Bhavesh
• Assistant Examiner(s): DESIRE, GREGORY M
• Application Number:
  US09/351,904
• Time in Patent Office: 1,385 Days
• Field of Search: 382/103, 382/104, 382/209, 382/278, 382/190, 382/218, 382/227, 382/302
• US Class Current: 382/104
• CPC Class Codes: G06V 10/757 Matching configurations of points or features
• CPC Class Codes: G06V 20/58 Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
• CPC Class Codes: G06V 20/582 of traffic signs

Abstract

An image-processing method, and an apparatus for carrying out the method, that is used particularly for detecting and recognizing objects in traffic. Locally-softened feature images of an image scene are correlated with feature images of patterns for detecting and recognizing objects in real scenes with the use of images. A plurality of patterns ordered in a pattern tree structure is correlated with the image scene. To extensively prevent erroneous detections, the features are subdivided into different feature types.

Claims

1. An image-processing method for correlating one or a plurality of feature image patterns with an original image, a feature image scene being formed from an original image scene through feature extraction, feature values of the feature image scene being locally softened through a transformation or a filter to define a transformed image scene, and the transformed image scene being correlated with the feature image patterns, and/or the feature values of the feature image patterns being softened through a transformation or a filter to define transformed feature image patterns, the transformed feature image patterns being correlated with the feature image scene, and with objects being detected and recognized with the processed image data and as a function of knowledge about the objects to be detected and/or on global knowledge about the situation in the scenario associated with the original image, the method comprising the steps of:

correlating N feature image patterns with the feature image scene through creation of a pattern hierarchy having a plurality of levels, the pattern hierarchy including prototype patterns that are formed from a group of feature image patterns from a hierarchically subordinate level; and

correlating the image scene with further prototype patterns from the hierarchically subordinate level in image regions in which a prototype pattern is successfully correlated with the feature image scene.

2. The image-processing method as recited in claim 1,

wherein the N feature image patterns of the objects to be detected and recognized are analytically generated with the use of samples or a priori knowledge about the objects.

3. The image-processing method as recited in claim 1,

wherein the correlation of the feature image scene using the pattern hierarchy is combined with a coarse-to-fine pixel grid sampling that is either as fine as or finer than the prototype pattern of the hierarchically subordinate level of the pattern hierarchy.

4. The image-processing method as recited in claim 1,

wherein the feature image scene and each of the N feature image patterns are subdivided into M different feature images according to feature type;

and

the M feature images of the image scene are correlated with M feature images of the patterns having a respective identical feature, and the correlation values are subsequently added.

5. The image-processing method as recited in claim 4,

wherein edges of an object are extracted as features from an image scene and the N patterns;

the edge image of the image scene and/or each of the edge images of the N patterns is or are subdivided into M edge images in corresponding intervals of edge orientations;

the feature values of the M edge images are locally softened;

the feature values of the corresponding edge image pattern are correlated with those of the transformed image scene for identical intervals; and

the correlation values of the intervals are subsequently added.

6. The image-processing method as recited in claim 1,

wherein patterns are only searched in an image cutout that is determined through the incorporation of global knowledge or assumptions about continuity of object motion, or information obtained via additional sensors or other pattern-processing methods.

7. The image-processing method as recited in claim 1 wherein objects in traffic are detected and recognized.

8. The image-processing method as recited in claim 7, wherein traffic infrastructure devices or road users are detected and recognized.

9. The image-processing method as recited in claim 1, wherein the detection results of the pattern-processing method are further processed in a regulating unit for controlling vehicle driving behavior and/or the flow of traffic.

10. The image-processing method as recited in claim 1,

wherein the correlation values serve in the selection of an image cutout in which patterns for an object are detected;

features are extracted in the image cutout, and a feature vector is determined; and

known classification methods are used to allocate the feature vector to a specific object class.

11. The image-processing method as recited in claim 1,

wherein the object positions determined in image coordinates with the detection and recognition results are transformed into 2D or 3D global coordinates.