Object detection apparatus, method and program

• US 8,121,348 B2
• Filed: 07/09/2007
• Issued: 02/21/2012
• Est. Priority Date: 07/10/2006
• Status: Active Grant

• Current Assignee: Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Original Assignee: Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Inventors: Hayasaka, Shoichi, Hara, Yoshimasa, Ichikawa, Naotaka, Fukamachi, Hideo, Sawaki, Toshio
• Primary Examiner(s): BHATNAGAR, ANAND P
• Application Number:
  US12/308,582
• Publication Number:
  US 20100226532A1
• Time in Patent Office: 1,688 Days
• Field of Search: 382/100, 382/103, 382/104, 382/106, 382/107, 382/181, 382/190, 382/191, 382/192, 382/195, 382/168, 382/170, 382/171, 348/135, 348/143, 348/148, 701/1, 701/23, 701/200, 701/300, 701/301
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 7/77 using statistical methods
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors
• 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: G08G 1/166 for active traffic, e.g. moving vehicles, pedestrians, bikes

Abstract

An object detection apparatus for detecting an object from an image obtained by taking a front view picture of a road in a traveling direction of a vehicle includes a camera unit for taking the front view picture of the road and inputting the image; a dictionary modeling the object; a search unit for searching the image with a search window; a histogram production unit for producing a histogram by comparing the image in the search window with the dictionary and counting a detection frequency in a direction parallel to a road plane; and a detection unit for detecting the detection object by detecting a unimodal distribution from the histogram.

Claims

1. An object detection apparatus that detects an object from an image obtained by taking a picture of a road before a vehicle in the direction of travel, the object detection apparatus comprising:

a camera unit that captures and inputs the image of the road before the vehicle;

a dictionary modeling the object;

a search unit that searches the image with a search window;

a histogram production unit that produces a histogram by comparing the image in the search window with the dictionary and counting a detection frequency in a direction parallel to a road plane; and

a detection unit that detects the object by detecting a unimodal distribution from the histogram.

2. The object detection apparatus according to claim 1, further comprising a determination unit that determines the detection of the object is erroneous if the histogram has a multimodal distribution.

3. The object detection apparatus according to claim 1, wherein:

the search unit searches the image while a size of the search window is changed;

the histogram production unit produces the histogram for each size of the search window; and

the object detection apparatus further comprises an object size estimation unit that estimates a size of the object based on a maximum detection frequency in the histogram for each size of the search window W produced by the histogram production unit.

4. The object detection apparatus according to claim 3, wherein the object size estimation unit estimates that the size of the search window whose maximum detection frequency is the greatest is a size of the object.

5. The object detection apparatus according to claim 1, wherein the dictionary has a resolution determined by searching a plurality of sample images containing the object with the search window, comparing the sample images in the search window with a plurality of dictionaries of different resolutions, counting a detection frequency in a direction parallel to a road plane to thereby produce histograms on a dictionary-by-dictionary basis, determining lowest resolutions among resolutions of the dictionaries that allow the histograms to have a unimodal distribution, and averaging the lowest resolutions derived with respect to the plurality of sample images.

6. The object detection apparatus according to claim 1, wherein the search unit searches the image with the search window in a predetermined pixel skipping amount, the predetermined pixel skipping amount is determined by searching a plurality of sample images containing the object with the search window, comparing the sample images in the search window with the dictionary, counting a detection frequency in a direction parallel to a road plane to thereby produce histograms, calculating maximum pixel skipping amounts in a direction perpendicular to a road plane based on a shape of the histograms, and averaging the maximum pixel skipping amounts derived for the plurality of sample images to obtain an average maximum pixel skipping value, which serves as an upper limit of the predetermined pixel skipping amount.

7. The object detection apparatus according to claim 6, wherein the maximum pixel skipping amount is a width of a peak in the normal distribution of the histogram.

8. The object detection apparatus according to claim 1, wherein the search unit searches the image in a predetermined search window size changing amount, the predetermined search window size changing amount being determined by searching a plurality of sample images m containing the object while a size of the search window is changed, comparing the sample images in the search window with the dictionary, counting a detection frequency in a direction parallel to a road plane to thereby produce histograms, calculating maximum search window size changing amount based on a maximum detection frequency in the histograms for each search window size, and averaging the maximum search window size changing amounts derived for the plurality of sample images m to obtain an average maximum search window size changing value that serves as an upper limit of the predetermined search window size changing amount.

9. The object detection apparatus according to claim 8, wherein a graph is produced such that one axis represents the size of the search window and the other axis represents the maximum detection frequency, and a width of a peak of the graph is determined as the maximum size changing amount.

10. The object detection apparatus according to claim 1, wherein the detection unit confines a search region to around an apex of the normal distribution, and specifies a position of the object.

11. The object detection apparatus according to claim 1, wherein if the search window and the dictionary differ in size, the size of the search window is changed to fit the size of the dictionary.

12. An object detecting method that detects an object from an image acquired by taking a picture of a road before a vehicle in the direction of travel, the method comprising:

acquiring an image of the road before the vehicle in the direction of travel and inputting the image;

searching the acquired image with a search window;

producing a histogram by comparing the image in the search window with a dictionary modeling the object and counting a detection frequency in a direction parallel to a road plane; and

detecting the object by detecting a unimodal distribution from the histogram.

13. A program storage device comprising;

a recording medium that is readable by an object detection apparatus for object detection that detects an object from an image acquired by taking a picture of a road before a vehicle in the direction of travel; and

a program code that is stored on the recording medium, and allows the object detection apparatus to function as;

a search unit that searches the image acquired by taking the front view picture of the road with a search window;

a histogram production unit that produces a histogram by comparing the image in the search window with a dictionary modeling the detection object and counting a detection frequency in a direction parallel to a road plane; and

a detection unit that detects the detection object by detecting a unimodal distribution from the histogram.