Correction of camera motion between two image frames

• US 5,751,838 A
• Filed: 01/26/1996
• Issued: 05/12/1998
• Est. Priority Date: 01/26/1996
• Status: Expired due to Term

• Current Assignee: NEC Corporation
• Original Assignee: NEC Research Institute, Inc. (NEC Corporation)
• Inventors: Roy, Sebastien, Cox, Ingemar J.
• Primary Examiner(s): NGUYEN, ANH-VINH THI
• Application Number:
  US08/592,249
• Time in Patent Office: 837 Days
• Field of Search: 382/107, 382/103, 382/236, 382/295, 382/296, 348/407, 348/413, 348/431, 348/571, 348/155
• US Class Current: 382/107
• CPC Class Codes: G06T 7/246 using feature-based methods, e.g. the tracking of corners or segments
• CPC Class Codes: G06V 10/147 Details of sensors, e.g. sensor lenses fingerprint or palmprint sensors G06V40/13; vascular sensors G06V40/145; eye sensors G06V40/19

Abstract

A technique for determining the ego-motion between two frames of an image that includes estimating likely values of the translation and rotation and making a five-dimensional search over the space of likely translations and rotations and finding the global minimum at the current motion estimate by a sum of squared differences cost function.

Claims

1. A process for determining ego-motion between two image frames of a scene that includes the steps of:

estimating the two parameters of translation and three parameters of rotation involved in camera ego-motion based upon pixels in one frame and corresponding epipolar line segments in the other frames;

computing the cost function of the estimated movement; and

revising the estimate and computing the cost function in iterative fashion to find the estimate that results in the cost function with the least figure of merit for use as the desired ego-motion.

2. The process of claim 1 in which the cost function depends on the sum of squared differences between pixels of one image and sample points along the assumed corresponding epipolar line segments in the other image.

3. The process of claim 1 that includes as a preliminary stop the normalization of the pixel intensity levels of the two frames.

4. The process of claim 3 in which the normalization is done by dynamic histogram warping.

5. A method for deriving a measure of the camera translation involved between two image frames of a given scene that have an assumed known rotation and an unknown translation comprising the steps of:

selecting a point A in the first frame;

computing the corresponding zero disparity point in the second frame;

selecting a plurality of different line segments that pass through the zero disparity point and are oriented as possible candidates for the focus of expansion;

selecting a number of points on each of the line segments;

calculating the error measures of each of the selected lines for finding the one with the minimum error; and

using the one with the minimum error for identifying the true focus of expansion.

7. The process of claim 5 that includes as a preliminary step the normalization of the pixel intensity levels of the two frames.

6. A method for deriving a measure of the camera rotation involved between two image frames of a given scene that have an assumed known translation and an unknown rotation comprising the steps of:

selecting a point A in the first frame;

computing the corresponding zero disparity point in the second frame;

selecting a plurality of different line segments that pass through the possible candidate zero disparity points and are oriented according to the focus of expansion;

selecting a member of points on each of the line segments;

calculating the error measures of each of the selected lines for finding the one with the minimum error; and

using the one with the minimum error for identifying the true focus of expansion.

8. The process of claim 6 that includes as a preliminary step the normalization of the pixel intensity levels of the two frames.