IMAGE PROCESSING APPARATUS AND IMAGE PROCESSING METHOD

• US 20150269449A1
• Filed: 12/16/2014
• Published: 09/24/2015
• Est. Priority Date: 03/24/2014
• Status: Active Grant

• Current Assignee: Alpine Electronics Incorporated (Alps Alpine Co., Ltd.), Toshiba Digital Solutions Corporation (Toshiba Corporation)
• Original Assignee: Toshiba Corporation
• Inventors: KOSAKI, Masanori
• Granted Patent: US 9,773,176 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: B60R 11/04 Mounting of cameras operative during drive; Arrangement of controls thereof relative to the vehicle
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06T 2207/30261 Obstacle
• CPC Class Codes: G06T 7/73 using feature-based methods
• 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 2201/07 Target detection
• CPC Class Codes: H04N 5/28 Mobile studios

Abstract

According to one embodiment, an image processing apparatus and an image processing method includes a processed image generation unit, a provisional detection unit, and a final detection unit. The processed image generation unit generates a processed image for detection processing from an image around a vehicle taken by a camera provided on the vehicle. The provisional detection unit scans a scanning frame of a predetermined size according to a detection target object on the processed image, and detects a plurality of position candidate regions of the detection target object within the processed image by determining a feature value for respective scanning positions using a dictionary of the detection target object. The final detection unit determines an overlapping region of the plurality of position candidate regions, and determines a final detection position of the detection target object within the processed image based on the overlapping region and the plurality of position candidate regions.

Claims

1. An image processing apparatus, comprising:

a processed image generation unit configured to acquire an image around a vehicle picked up by a camera provided on the vehicle, and generate a processed image from the image for target object detection processing;

a provisional detection unit configured to scan a scanning frame of a predetermined size according to a detection target object on the processed image, and detect a plurality of position candidate regions of the detection target object within the processed image by determining a feature value for respective scanning positions using a dictionary of the detection target object; and

a final detection unit configured to determine an overlapping region of the plurality of position candidate regions, and determine a final detection position of the detection target object within the processed image based on the overlapping region and the plurality of position candidate regions.

2. The apparatus according to claim 1, wherein:

the final detection unit determines the final detection position of the detection target object within the processed image based on a history of overlapping regions in a current frame image and a past plurality of frame images, and the plurality of position candidate regions of the current frame.

3. The apparatus according to claim 1, wherein:

the provisional detection unit determines a likelihood using the dictionary for each scanning position of the scanning frame of the predetermined size, and detects scanning position for which a likelihood is greater than or equal to a predetermined likelihood as the plurality of position candidate regions; and

the final detection unit extracts a plurality of valid position candidate regions from the plurality of position candidate regions of a current frame image based on a history of overlapping regions in a current frame and a past plurality of frames, and determines the final detection position by subjecting the plurality of valid position candidate regions to weighted-addition averaging using respective likelihoods of the plurality of valid position candidate regions.

4. The apparatus according to claim 3, wherein:

the provisional detection unit changes the predetermined likelihood for each scanning position of the scanning frame of the predetermined size based on the history of the overlapping regions.

5. The apparatus according to claim 1, wherein:

the processed image generation unit generates a plurality of processed images by scaling the image picked up the camera so as to cause the plurality of processed images to have a different scaling factor to each other while each of the plurality of processed images having a predetermined visual axis as a normal-line direction;

the provisional detection unit classifies the plurality of processed images into a plurality of categories according to the scaling factor, and detects the position candidate regions for each category; and

the final detection unit determines the overlapping region, for each category, of the plurality of position candidate regions, and determines the final detection position based on histories of overlapping regions in a current frame image and a past plurality of frame images, each history being corresponding to each category.

6. The apparatus according to claim 1, wherein:

the final detection unit corrects the overlapping region using an optical flow between an immediately preceding frame image and a current frame image, and determines the final detection position using corrected overlapping region.

7. The apparatus according to claim 1, wherein:

the final detection unit changes a shape of the plurality of position candidate regions detected by the provisional detection unit in accordance with a type of the detection target object, and thereafter determines the overlapping region.

8. The apparatus according to claim 1, further comprising:

a dictionary generation unit configured to generate the dictionary as a single dictionary in advance using a plurality of images of the detection target object picked up in advance by the camera, the plurality of images being obtained by arranging the detection target object at a corresponding plurality of positions in a predetermined range within an image-pickup range of the camera so as to make a predetermined visual axis of the camera and a normal-line direction of a detection target surface of the detection target object parallel to each other and being obtained by picking up each of the plurality of images by the camera at the corresponding plurality of positions,

wherein the processed image generation unit generates as the processed image an image having the predetermined visual axis as a normal-line direction, based on an image that is picked up by the camera after generation of the dictionary.

9. The apparatus according to claim 8, wherein:

the provisional detection unit detects the plurality of position candidate regions using the dictionary based on a HOG feature value.

10. The apparatus according to claim 1, further comprising:

a warning unit configured to notify a driver of the vehicle of information regarding a final detection position that is detected by the final detection unit using at least one of an audio output through a speaker of the vehicle, a buzzer output through the speaker of the vehicle, and a warning display on a display apparatus provided at a position recognizable to the driver.

11. An image processing method comprising:

acquiring an image around a vehicle picked up by a camera provided on the vehicle;

generating a processed image from the image for target object detection processing;

scanning a scanning frame of a predetermined size according to a detection target object on the processed image;

detecting a plurality of position candidate regions of the detection target object within the processed image by determining a feature value for respective scanning positions using a dictionary of the detection target object;

determining an overlapping region of the plurality of position candidate regions; and

determining a final detection position of the detection target object within the processed image based on the overlapping region and the plurality of position candidate regions.

12. A non-transitory computer readable medium having instructions stored therein causes, when executed by a computer, to execute the steps of the method according to claim 11.