WO2012164804A1 - 物体検出装置、物体検出方法および物体検出プログラム - Google Patents
物体検出装置、物体検出方法および物体検出プログラム Download PDFInfo
- Publication number
- WO2012164804A1 WO2012164804A1 PCT/JP2012/002375 JP2012002375W WO2012164804A1 WO 2012164804 A1 WO2012164804 A1 WO 2012164804A1 JP 2012002375 W JP2012002375 W JP 2012002375W WO 2012164804 A1 WO2012164804 A1 WO 2012164804A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- video
- distance
- object detection
- image
- unit
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
- G06V10/7515—Shifting the patterns to accommodate for positional errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
Definitions
- the present invention relates to an object detection device, an object detection method, and an object detection program, and more particularly, to an object detection device, an object detection method, and an object detection program capable of detecting an object such as a pedestrian existing around a vehicle.
- an in-vehicle camera imaging device
- a front part or a rear part of a vehicle captures an image around the vehicle.
- the captured video is displayed on a display installed in the driver's seat.
- the driver can confirm the presence or absence of an object such as a pedestrian around the vehicle when the vehicle is traveling.
- a sensor that detects a heat source and a distance is mounted separately from the in-vehicle camera.
- the mounting of another sensor has disadvantages in terms of cost and versatility, and there is a demand for practical use of an image recognition means that detects an object only from an on-vehicle camera image.
- the above-described object detection method based on image recognition still has a problem related to processing time.
- image recognition since video processing is complicated and it is necessary to scan the entire video, it takes time to obtain an object detection result in one frame of video. For this reason, there has been a problem that the frame rate of the object detection process is lower than the frame rate of the camera and a delay time until an object detection result is obtained for the video.
- the object detection device described in Patent Document 1 creates a reduced image obtained by reducing the input image, first detects roughly the presence or absence of an object on the reduced image, and displays the reduced image on the reduced image.
- Discloses a technique for performing two-stage detection in which detection processing is performed again with an input image of the original size only when an object is detected in (1). Accordingly, scanning is performed with a reduced image size, so that the detection process can be speeded up.
- An object of the present invention is to provide an object detection apparatus, method, and program capable of speeding up detection processing time without sacrificing detection performance while performing object detection only by image recognition. .
- One aspect of the present invention is an object detection device that detects an object in the vicinity of a vehicle from an input image using an image around the vehicle captured from the vehicle as an input image.
- the object detection device converts an input image into an input image.
- the video conversion unit that converts the image features from the image to the extracted feature video, and extracts the different regions for each distance from the feature video as a video by distance, and the composite video using the video by distance
- the first object detection process is performed to scan the synthesized video synthesized by the distance-based video extraction and synthesis unit and the distance-based video extraction and synthesis unit to obtain the position of the object detected from the synthesized video on the synthesized video.
- a first object detection unit, and an object candidate position designation unit that obtains an object candidate position where an object may exist in the image by distance based on the position of the object detected by the first object detection unit on the composite image;
- Object candidate A second object detection unit that performs a second object detection process for identifying a corresponding object position in the image by distance with respect to the object candidate position obtained by the position designation unit, and the object position obtained by the second object detection unit
- an object position specifying unit for specifying an object position in the input video.
- Another aspect of the present invention is an object detection method for detecting an object in the vicinity of a vehicle from an input image using an image around the vehicle photographed from the vehicle as an input image, and the object detection method uses an input image as an input image.
- the video conversion step that converts the image features into the feature video extracted from the image, the different video for each distance is extracted from the feature video as the distance video, and the composite video using the distance video.
- the first object detection processing is performed to obtain the position of the object detected from the synthesized video by scanning the synthesized video synthesized by the distance-based video extracting and synthesizing step and the distance-based video extracting and synthesizing step.
- an object candidate position where the object may exist in the image by distance is obtained.
- Another aspect of the present invention is an object detection program for detecting an object in the vicinity of a vehicle from an input image using an image around the vehicle photographed from the vehicle as an input image.
- the object detection program sends an input image to a computer.
- Video conversion function that converts image features from input video into feature video, and distance-based video extraction that extracts and synthesizes different areas for each distance from feature video based on distance from vehicle
- the first object detection function that performs the first object detection process by scanning the composite video that is synthesized by the synthesis function and the distance-based video extraction and synthesis function, and the position on the composite video where the object is detected by the first object detection function
- a second object detection function for performing object detection processing and specifying an object position in a video according to distance, and an object position specifying function for specifying an object position in an input video based on the object position obtained by
- FIG. 1 is a block diagram showing a configuration of an object detection apparatus according to an embodiment of the present invention.
- 2 (a) is an explanatory diagram showing an example of a pedestrian having a height of 200cm at a distance Am point.
- FIG. 2 (b) is an explanatory diagram showing an example of a pedestrian having a height of 200cm at a point of distance Bm.
- (C) is an explanatory diagram showing an example in which a pedestrian having a height of 200 cm is present at a distance Cm.
- FIG. 2D is an explanatory diagram showing an example in which a pedestrian having a height of 100 cm is present at a distance Am.
- FIG. 3A is an explanatory diagram showing an imaging region when a pedestrian having a height of 200 cm exists at a distance Am point.
- FIG. 3B is an imaging region when a pedestrian having a height of 200 cm exists at a point of distance Bm.
- 3 (c) is an explanatory diagram showing an imaging region when a pedestrian having a height of 200 cm is present at a distance Cm.
- FIG. 3 (d) is an explanatory diagram showing a pedestrian having a height of 100 cm at a distance Am.
- FIG. 3 (e) is an explanatory diagram showing an imaging region when a pedestrian having a height of 100 cm is present at a distance Bm.
- FIG. 3 (f) is an explanatory diagram showing a pedestrian having a height of 100 cm. Explanatory drawing which shows the imaging region when existing at a point
- FIG. 4 is an explanatory diagram for comparing imaging regions when a pedestrian having a height of 200 cm and 100 cm exists at a point of a distance
- Am. 5 (a) is an explanatory diagram showing an example of an Am video corresponding to the Am point.
- FIG. 5 (b) is an explanatory diagram showing an example of a Bm video corresponding to the Bm point.
- FIG. 5 (c) is a diagram showing the Cm point.
- FIG. 6A is an explanatory diagram showing an example before and after the Am video is scaled
- FIG. 6B is an explanatory diagram showing an example before and after the Bm video is scaled.
- FIG. 7 is an explanatory diagram showing an example of the center alignment of the video by distance.
- FIG. 8 is an explanatory diagram showing an example of the number of overlapping images by distance.
- 9A is a diagram showing an example of the composite number of Am video and an image feature amount.
- FIG. 9B is a diagram showing an example of the composite number of Bm video and an image feature amount.
- FIG. 9C is a diagram showing Cm.
- FIG. 10 (a) is a diagram showing an example of a composite coefficient for Am video.
- Fig. 10 (b) is a diagram showing an example of a composite coefficient for Bm video.
- Fig. 10 (c) is a diagram showing an example of a composite coefficient for Cm video.
- Figure 11 (a) is a diagram showing an example of an input video
- FIG. 11 (b) is a diagram showing an example of an Am video
- FIG. 11 (c) is a diagram showing an example of a Bm video
- FIG. 11 (d) is Cm.
- Fig. 11 (e) is a diagram showing an example of a composite image.
- FIG. 12 is an explanatory diagram showing an example of a pedestrian correctly normalized on the composite video
- FIG. 13 (a) is an explanatory diagram showing an example of clipping from a synthesized video targeting a height of 100 cm.
- FIG. 13 (b) is an explanatory diagram showing an example of enlarging the synthesized video after clipping.
- FIG. 14 is a diagram illustrating an example of image feature count scanning on a composite video.
- FIG. 15A is an explanatory diagram showing an example of the input video.
- FIG. 15B is an explanatory diagram showing an example of the detection result obtained by performing the first part object detection process on the synthesized video.
- FIG. An explanatory diagram showing an example of specifying an object candidate position on the Am video.
- FIG. 15D is an explanatory diagram showing an example of specifying an object candidate position on the Bm video.
- FIG. 15E is an object candidate on the Cm video.
- Explanatory drawing showing an example of specifying the position 16 (a) is an explanatory diagram showing an example of an object detection result in an Am video.
- FIG. 16 (b) is an explanatory diagram showing an example of an object detection result in a Bm video.
- FIG. 16 (c) is an object detection in a Cm video.
- FIG. 17A is an explanatory diagram showing an example of the extracted coordinates of the video by distance on the input video.
- FIG. 17B is an explanatory diagram showing an example of the detected coordinates on the video by distance.
- FIG. Is an explanatory diagram showing an example of detected coordinates on the input video
- An object detection device is an object detection device that detects an object in the vicinity of a vehicle from an input image using an image around the vehicle captured from the vehicle as an input image, and extracts the input image and image features from the input image.
- a video conversion unit that converts to a feature video, and a video by distance that extracts different areas as distance-based video from the feature video based on the distance from the vehicle and synthesizes a composite video using the video by distance
- a first object detection unit that performs a first object detection process for obtaining a position on the composite image of an object detected from the composite image by scanning the composite image synthesized by the extraction and synthesis unit; ,
- An object candidate position specifying unit for obtaining an object candidate position where an object may exist in the image according to distance based on the position of the object detected by the first object detection unit, and an object candidate position specifying unit Asked in
- a second object detection unit that performs a second object detection process for identifying a corresponding object position in the image according to distance with respect to the object candidate
- the video conversion unit extracts an edge feature as an image feature.
- the input video can be converted into a video that retains only the edge features used in the object detection process.
- the image extraction / synthesis unit for each distance changes the size of the region extracted from the feature image based on the distance from the vehicle.
- object detection can be performed in accordance with the size of the detection target object imaged as a different size on the video depending on the distance from the vehicle.
- the distance-by-distance video extraction / synthesizing unit scales the video extracted for each distance so that the vertical sizes of all the distance-by-distance videos are equal.
- the distance-by-distance video extraction / synthesizing unit synthesizes the positions of the horizontal central axis and the vertical central axis of the distance-by-distance video.
- the center position of the composite image is aligned for all the distance-based images, so that the position of the detected object can be easily grasped.
- the size of the composite image is smaller than the total size of all the distance-based images, the object detection process can be speeded up.
- the distance-by-distance video extraction and synthesis unit synthesizes the distance-by-distance video by ⁇ blending.
- the synthesis coefficient it is possible to adjust the degree of influence of each distance-based video in the synthesized video.
- the distance-by-distance video extraction / synthesis unit partially synthesizes by adjusting the ⁇ blending synthesis coefficient partially in accordance with the number of synthesized videos by distance.
- the distance-by-distance video extraction / synthesis unit adjusts the ⁇ blending synthesis coefficient in accordance with the image feature amount included in the distance-by-distance video.
- the vertical size of the object to be detected in the object detection process of the first object detection unit is equal to the vertical size of the composite image.
- the first object detection unit performs the object detection process by scanning the composite image only in the horizontal direction. With this configuration, since the number of object detection scans can be reduced, the object detection process can be speeded up.
- the first object detection unit cuts out an area in contact with the lower end of the composite video and expands the vertical size of the cut-out area so as to be equal to the vertical size of the composite video
- a synthesized video cutout enlargement unit that generates With this configuration, it is possible to use the assumption that the lower end of the object is always in contact with the lower end of the composite video when dealing with differences in the size of the detection target object (for example, differences in pedestrian height). Since the region on the video where no object can exist is not scanned, the object detection process can be speeded up. Further, since the size of the cut out synthesized video is matched with the size of the synthesized video before being cut out, there is an advantage that it is not necessary to consider the size of the detection target object on the video in the object detection processing.
- the first object detection unit performs object detection processing on the enlarged composite image.
- a difference in size of the detection target object for example, a difference in pedestrian height
- the first object detection unit performs the object detection process by scanning the enlarged composite image only in the horizontal direction.
- the first object detection unit can adjust the detection criterion for performing object detection separately from the second object detection unit, and the object detection unit is more object-oriented than the second object detection unit. It is preferable that the detection reference is adjusted so that it can be easily determined that there is. With this configuration, the first object detection unit makes coarse detection, that is, it is easy to detect, prevents detection omission on the composite image, and the second object detection unit strictly determines that the detection object exists. , Can prevent false detection.
- the first object detection unit performs the first object detection process only for a portion where an image feature exists on the synthesized video. With this configuration, it is not necessary to scan a portion where a detection object cannot exist, so that the detection process can be speeded up.
- the object position specifying unit obtains the distance from the vehicle to the detected object based on the object position on the image according to distance detected by the second object detection unit.
- An object detection method of the present invention is an object detection method for detecting an object in the vicinity of a vehicle from an input image using an image around the vehicle photographed from the vehicle as an input image, and extracting an input image and an image feature from the input image Based on the video conversion step to convert to feature video and the distance from the vehicle, a different video for each distance is extracted from the feature video as a video by distance, and the video by distance that synthesizes the composite video using the video by distance
- a first object detection step for performing a first object detection process for obtaining a position on the composite video of an object detected from the composite video by scanning the composite video synthesized in the extraction video synthesis step and the distance-based video extraction and synthesis step; Detecting object candidate position for finding object candidate position where object may exist in video according to distance based on position of synthesized object detected in first object detection step And a second object detection step for performing a second object detection process for identifying a corresponding object position in the image by distance with respect to the object candidate position obtained in the object candidate position detection step,
- An object detection program of the present invention is an object detection program for detecting an object near a vehicle from an input image using an image around the vehicle photographed from the vehicle as an input image, wherein the input image is imaged from the input image to a computer.
- a video conversion function that converts images into feature video
- a distance-based video extraction and synthesis function that extracts and synthesizes different regions for each distance from the feature video based on distance from the vehicle, and distance-specific video
- Based on the first object detection function that scans the synthesized video synthesized by the video extraction and synthesis function and performs the first object detection process, and the position on the synthesized video where the object was detected by the first object detection function
- An object candidate position detection function that obtains an object candidate position where an object may exist in the image, and a second object detection process for the object candidate position obtained by the object candidate position detection function
- a second object detection function for specifying an object position in a video according to distance
- an object position specification function for specifying an object position in an input video based
- the detection processing time can be increased without sacrificing the detection performance, and the distance from the vehicle to the object can also be measured. It is possible to obtain an excellent effect that is possible.
- FIG. 1 is a block diagram showing a configuration of an object detection apparatus according to an embodiment of the present invention.
- the object detection apparatus 10 shown in the figure converts an input video input from the outside into a feature video obtained by extracting an image feature from the input video, and from the feature video for each distance based on the distance from the vehicle.
- a distance-by-distance video extracting and synthesizing unit 30 for synthesizing a video by distance from which different areas are extracted, a first object detecting unit 40 for performing a first object detection process from the obtained synthesized video, and a result of the first object detection process
- An object candidate position specifying unit 50 for obtaining an object candidate position where an object may exist in a distance-by-distance video
- a second object detection unit for performing a second object detection process on the object candidate position on the distance-by-distance video 60 and an object position specifying unit 70 for specifying an object position on the input video from the detection result of the second object detecting unit.
- the input image input from the outside is, for example, an image obtained by photographing the periphery of the vehicle with a vehicle-mounted camera attached at a predetermined angle near a license plate on the front or rear side of the vehicle.
- the video conversion unit 20 performs video conversion processing for extracting image features on the video based on the input video.
- the edge feature is extracted as the image feature, and the processing focusing on the edge feature is performed in the subsequent processing.
- the target image feature in the present invention is limited to the edge feature. It is not a thing. For example, for the purpose of detecting signs and traffic lights on the road, it is effective to perform a process of extracting a specific color as an image feature.
- Specific processing for extracting edge features includes embossing and edge extraction using a Sobel filter. However, the present invention is not limited to these processes.
- the distance-by-distance video extraction / composition unit 30 includes a distance-by-distance video extraction unit 31, a distance-by-distance video enlargement / reduction unit 32, a center alignment unit 33, a distance-by-distance video feature amount determination unit 34, a synthesis coefficient adjustment unit 35, Part 36 is provided.
- the distance-by-distance video extraction unit 31 has a different size for each distance in a region where the detection target object may be imaged on the video when the detection target object is located at a predetermined distance from the vehicle. , Extracted from each feature video.
- the distance-by-distance video enlargement / reduction unit 32 enlarges or reduces the distance-by-distance video corresponding to each distance extracted by the distance-by-distance video extraction unit 31 so that the vertical size of the distance-by-distance video becomes a predetermined size.
- the vertical sizes of all the images by distance are equal (the horizontal sizes are different).
- the video by distance refers to the video by distance after being enlarged or reduced by the video enlargement / reduction unit 32 by distance.
- the center position aligning unit 33 aligns the positions of the horizontal center axis and the vertical center axis of the distance-based image resized by the distance-based image enlargement / reduction unit 32 and obtains the number of composites based on the image position.
- the number of composites is the number of superimposed images by distance.
- the distance-by-distance video feature determination unit 34 obtains an image feature amount existing on each distance-by-distance video.
- the synthesis coefficient adjustment unit 35 obtains a synthesis coefficient for each partial region of the video by distance based on the number of synthesis obtained by the center alignment unit 33 and the image feature quantity obtained by the video feature quantity determination unit 34 by distance.
- the distance-by-distance video synthesis unit 36 multiplies the distance-by-distance video obtained by adjusting the center position by the center alignment unit 33 by the synthesis coefficient obtained by the synthesis coefficient adjustment unit 35 to generate a synthesized video by synthesizing all the video by distance. To do. At this time, the synthesized video is smaller than the total number of pixels of all the video by distance.
- the first object detection unit 40 includes a composite video cutout enlargement unit 41, a composite video feature amount determination unit 42, and a composite video object detection unit 43.
- the synthesized video cut-out enlargement unit 41 cuts out a part of the synthesized video and enlarges the video so that the cut-out synthesized video becomes equal to the vertical size before being cut out.
- the purpose of this process is a process for dealing with a difference in the size of the object detection target (for example, a difference in pedestrian height).
- the composite video feature amount determination unit 42 specifies a location where an image feature exists on the composite video output from the composite video cut-out enlargement unit 41.
- the composite video object detection unit 43 performs object detection processing on only the portion where the image feature exists in the composite video feature amount determination unit 42 on the composite video output from the video cutout enlargement / magnification unit 41. Since there is no detection target object in the first place where there is no image feature, it can be expected to speed up the object detection process by excluding it in advance.
- the synthesized image in the first object detection unit and the subsequent stage refers to a synthesized image after cutting and enlargement.
- the object candidate position designating unit 50 determines where the detected position on the composite image detected by the first object detecting unit 40 corresponds to the image classified by distance. For example, if a composite video is combining two video images according to distance, if one object is detected on the composite video, there is one position where the object may exist on each video according to distance. Therefore, there are two object candidate positions in total.
- the second object detection unit 60 performs object detection only on the candidate positions on the distance-by-distance video designated by the object candidate position designation unit 50, and specifies the object positions existing on the distance-by-distance video.
- the object position specifying unit 70 calculates the object position on the input image based on the detection result of the second object detection unit 60, and further outputs the final result together with the distance from the vehicle to the object. .
- the input image is 640 pixels in the horizontal direction and 480 pixels in the vertical direction.
- the vertical size of the distance-by-distance video output from the distance-by-distance video enlargement / reduction unit 32 and the synthesized video output from the synthesized video cut-out enlargement unit 41 is 128 pixels.
- the size of the scanning frame when the target object detection processing is performed by the first object detection unit and the second object detection unit is assumed to be 64 pixels horizontally and 128 pixels vertically.
- the object to be detected is a pedestrian existing on the road surface, and the height of the pedestrian to be detected is 100 cm to 200 cm.
- the pedestrian who exists in three points, the distance from a vehicle, A meter, B meter, and C meter is made into a detection target. Note that the example given here is for explanation purposes, and the present embodiment is not limited to this.
- the distance-by-distance video extraction unit 31 generates a distance-by-distance video by extracting video from the input image for each distance from the vehicle.
- the pedestrian is imaged on the input video according to the distance between the vehicle and the pedestrian.
- the area to be determined is uniquely determined.
- FIG. 2 shows an input image when a pedestrian having a height of 200 cm and a pedestrian having a height of 100 cm are present at points of A meter, B meter, and C meter (hereinafter referred to as Am, Bm, and Cm) from the vehicle. .
- A, B, and C The magnitude relationship between A, B, and C is A ⁇ B ⁇ C, where Am is the closest to the vehicle and Cm is the farthest from the vehicle.
- a pedestrian having a height of 200 cm is present at a point at a distance Am in FIG. 2A
- FIG. 2B is at a point at a distance Bm
- FIG. 2C is at a point at a distance Cm.
- FIG. 2 (d) shows that a pedestrian with a height of 100 cm exists at a distance Am
- FIG. 2 (e) shows that a pedestrian exists at a distance Bm
- FIG. 3 shows a region where a pedestrian to be detected is imaged on each distance video when the input video is as shown in FIG.
- FIGS. 3A to 3F correspond to FIGS.
- the distance-by-distance video enlargement / reduction unit 32 enlarges / reduces each distance-by-distance video extracted by the distance-by-distance video extraction unit 31 and normalizes the vertical size of the distance-by-distance video to 128 pixels.
- the aspect ratio of the image by distance is unchanged before and after the enlargement / reduction.
- FIG. 6 shows an example of a distance-by-distance video enlargement / reduction process.
- FIG. 6A shows a distance-by-distance video corresponding to the distance Am (hereinafter referred to as Am video), and
- FIG. 6B shows a distance Bm.
- Corresponding video by distance hereinafter referred to as “Bm video”
- Cm video are before and after scaling of the video by distance corresponding to distance Cm (hereinafter referred to as “Cm video”).
- the horizontal size of each distance image before scaling is equal to 640 pixels, and the vertical size is Ya pixels for Am video, Yb pixels for Bm video, and Yc pixels for Cm video.
- the vertical size of the images by distance after scaling is 128 pixels, and the horizontal size is (640 ⁇ 120 / Ya) pixels for Am video, (640 ⁇ 128 / Yb) pixels for Bm video, and (640 ⁇ 128) for Cm video. / Yc) pixel.
- the sizes of the pedestrians are equal in the video according to distance after expansion / reduction regardless of the distance from the vehicle.
- the center position aligning unit 33 aligns the center position of the distance-by-distance video whose vertical size is normalized to 128 pixels by the distance-by-distance image enlargement / reduction unit 32.
- FIG. 7 shows an example in which three center positions of an Am video, a Bm video, and a Cm video are combined as the video by distance. As shown in FIG. 7, when the center positions of the distance-based images are matched, the vertical size of the distance-based images is normalized to 128 pixels, but the horizontal sizes are different. .
- FIG. 8 illustrates the number of overlapping images according to distance in the example of FIG. 7 and decreases from the center position to 3, 2, and 1. This number of overlaps is the composite number.
- the distance-by-distance video feature amount determination unit 34 counts how many image features each exist on the distance-by-distance video output from the distance-by-distance video enlargement / reduction unit 32.
- edge features are used as image features.
- the image feature amount included in the Am-by-distance video is Ea
- the image feature amount of Bm is Eb
- the image feature amount of Cm is Ec.
- edge features are targeted, but other image features can also be targeted.
- the presence / absence of a specific color may be determined, or processing for determining the presence / absence of a certain level of brightness or more may be performed.
- the distance-by-distance video output from the distance-by-distance video enlargement / reduction unit has a different video size. Therefore, the image feature amount existing in the video is normalized by the video size. May be.
- the synthesis coefficient adjustment unit 35 is based on the number of synthesized videos by distance obtained by the center alignment unit 33 and the image feature values Ea, Eb, and Ec for each video by distance obtained by the video feature value judgment unit 34 by distance. Then, a composite coefficient of each distance video is calculated.
- 9A to 9C summarize a correspondence list between the number of synthesized images and image feature amounts for the distance-by-distance videos of the distances Am, Bm, and Cm. A method for calculating the synthesis coefficient in such a case will be described with reference to FIG.
- FIGS. 10 (a) to 10 (c) show synthesis coefficients set for distance images of distances Am, Bm, and Cm.
- Am video is one type of synthesis coefficient Ma1
- Bm video is a synthesis coefficient Mb1.
- Two types of Mb2 and Cm video have three types of synthesis coefficients Mc1 to Mc3.
- Ma1, Mb1, and Mc1 are portions where three images of Am video, Bm video, and Cm video are combined
- Mb2 and Mc2 are portions where two images of Bm video and Cm video are combined.
- Mc3 is only one Cm video.
- the basic idea of the synthesis coefficient is that the synthesis coefficient is evenly allocated to the video for each distance to be synthesized according to the number of synthesis. That is, the total of three composite portions is 1 with 1/3 of each distance video, and the total of 2 composite portions is 1 with 1/2 of each distance video.
- the synthesis coefficient is adjusted using image feature amounts Ea to Ec corresponding to the distance videos.
- the idea of adjustment based on the image feature amount is to increase the composition coefficient of the distance-by-distance video having a large amount of image feature amount and to reduce the synthesis coefficient of the distance-by-distance image having a small image feature amount.
- the synthesis coefficients Ma1, Mb1, Mb2, Mc1, Mc2, and Mc3 can be obtained by the following formulas 1 to 6, respectively.
- calculation method is not limited to the above calculation formula as long as the total sum of the synthesis coefficients is 1.
- the distance-by-distance video synthesis unit 36 synthesizes the distance-by-distance video that has been aligned by the center alignment unit 33 using the synthesis coefficient obtained by the synthesis coefficient adjustment unit 35 to generate one synthesized video.
- General image blending is used for video composition processing using a composition coefficient.
- FIG. 11 is a diagram illustrating a specific example of video composition.
- FIG. 11A shows an example of the input video, and there are four pedestrians in the video. Explaining four pedestrians, the left part of the input image has two bodies of 200 cm and 100 cm tall at a distance of Am from the vehicle, and the center part of the input image has a height of 200 cm at a distance of Bm from the vehicle.
- FIG. 11B is an Am video
- FIG. 11C is a Bm video
- FIG. 11D is a Cm video
- FIG. 11E is a composite image obtained by normalizing the sizes of three images according to distance. It is an example of a result.
- the synthesized video cutout enlargement unit 41 cuts out a part of the synthesized video output from the distance-based video synthesis unit 36 and enlarges it to a predetermined size.
- the purpose of cutting out the synthesized video is to cope with the difference in the size of the detection target object. For example, in FIG. 11, it is for detecting both a 100 cm tall pedestrian and a 200 cm tall pedestrian.
- FIG. 12 shows a diagram in which only four pedestrians whose sizes are correctly normalized in accordance with the distance from the vehicle are extracted from the synthesized video in FIG. 11 (e).
- the positions of the feet of all pedestrians are the same regardless of the height, but the position of the head differs depending on the height, and if the height is the same, the size is the same regardless of the distance from the vehicle.
- the vertical size of the pedestrian needs to be about 128 pixels because of the setting of the scanning frame. Therefore, a pedestrian having a height of 100 cm cannot be detected as it is. Therefore, this is dealt with by cutting out and expanding the synthesized video.
- the image is enlarged while the aspect ratio of the image is preserved so that the extracted image has a vertical size of 128 pixels which is the vertical size of the synthesized image before being extracted.
- FIG. 13 is an example of cut-out enlargement for a height of 100 cm
- FIG. 13 (a) is a view showing a cut-out area
- FIG. 13 (b) is an enlarged view of the cut-out video.
- FIG. 13 (a) is a view showing a cut-out area
- FIG. 13 (b) is an enlarged view of the cut-out video.
- an example of cut-out enlargement with a height of 100 cm is shown, but cut-out enlargement processing is performed for other heights as much as necessary.
- the foot since the foot always exists at the lower end of the composite image regardless of the height of the pedestrian, the cutout position is always in contact with the lower end of the composite image.
- the cut-out enlargement process may be appropriately performed on the height to be detected.
- the composite video feature amount determination unit 42 will be described with reference to FIG.
- the image feature amount existing in the scanning frame (horizontal 64 pixels, vertical 128 pixels) of the object detection process on the composite video clipped and magnified by the composite video cropping enlargement unit 41 is counted.
- the edge feature is targeted, and the edge counting method is the same as that of the distance-by-distance video feature amount determination unit 34, and thus the description thereof is omitted.
- the composite video object detection unit 43 performs object detection processing on the composite video. Since the vertical size of the composite image and the vertical size of the scanning frame of the object detection process are the same, a process of scanning only once in the horizontal direction on the composite image and determining whether a detection target object exists in the scan Become. At this time, instead of performing detection processing at all scanning positions in scanning, it is determined whether to perform detection processing using the result of the composite video feature amount determination unit 42. That is, when there is no image feature in the scanning frame, no object can exist, and thus the entire detection process is speeded up by performing scanning without performing the detection process. Whether or not an image feature exists is determined by performing a detection process only when a threshold value is appropriately set and an image feature exceeding a certain level is present.
- the object detection process in this embodiment uses an object detection method based on edge features.
- a method using the edge feature a method using the HOG feature using the strength and direction of the edge is general and can be applied to the present invention. Further, the present invention is not limited to the method using the HOG feature, and any object detection using an edge feature can be applied. In the present invention, an image feature other than an edge feature can be applied.
- FIG. 15A shows the input video
- FIG. 15B shows the detection result of the first object detection process on the composite video
- FIG. 15C shows the output of the video enlargement / reduction unit 32 by distance from the first object detection process.
- FIG. 15D shows the result of designating the object candidate position on the Bm video
- FIG. 15E shows the result of designating the object candidate position on the Cm video.
- the input video here indicates that there are one pedestrian having a height of 200 cm at points of distance Am and Cm, and two pedestrians are detected as a detection result on the composite video.
- the synthesized video is composed of three videos of Am video, Bm video, and Cm video
- object candidates are present in each of the Am video, Bm video, and Cm video.
- One position can be specified. Since the center position of the synthesized video and each distance-based video match, the candidate position on each distance-based video can be designated on the same coordinates from the detection position on the synthesized video.
- the composite video is not composed of three videos in all areas, but is partially generated from two videos or one video, so depending on the position of the detection result on the composite video, it may not necessarily be Object candidate positions cannot be specified on all distance-based images. In the example shown in FIG. 15, one object candidate position is designated on the Am video, and two object candidate positions are designated on the Bm and Cm distance-by-distance videos.
- the second object detection unit 60 performs object detection processing on the object candidate positions designated on the video for each distance by the object candidate position designation unit 50, and specifies the position where the object exists on the video for each distance.
- the object detection process here does not need to be scanned as performed by the composite video object detection unit 43, and the object detection process may be limited to the candidate positions designated by the object candidate position designation unit 50.
- FIG. 16 shows detection results obtained by performing the object detection process on the object candidate positions designated in FIGS. 15 (c) to 15 (e).
- 16A shows detection results in Am video
- FIG. 16B shows detection results in Bm video
- FIG. 16C shows detection results in Cm video, and 1 in Am video and Cm video. The body is detected.
- the object detection processing means an object detection method based on edge features is used as in the composite video object detection unit 43.
- the method used may be the same detection method as that of the composite video object detection unit 43 or may be different.
- the accuracy of object detection in the composite video object detection unit 43 and the second object detection unit 60 may be changed. In this case, the detection accuracy of the second object detection unit 60 may be made stricter than the synthesized video object detection unit 43.
- the composite video object detection unit 43 Since the composite video object detection unit 43 has a larger processing amount than the second object detection unit 60, it performs simple object detection to the extent that some erroneous detection is allowed, and the object candidates are limited to reduce the processing amount. In the second object detection process, erroneous detection is excluded, and only the detection target object is reliably set as a detection result. As a result, it is possible to increase the speed of the object detection process without reducing the detection accuracy.
- the object position specifying unit 70 specifies the object position on the input image based on the object detection result on the image by distance detected by the second object detection unit 60.
- the object position on the input video can be easily obtained from the extracted coordinate position and the enlargement / reduction ratio when the video by distance is generated by the video extraction unit 31 by distance and the video enlargement / reduction unit 32 by distance.
- a procedure for obtaining the position coordinates on the input image from the detected position coordinates on the image by distance will be described with reference to FIG.
- FIG. 17A shows the coordinate position for extracting video by distance from the input video.
- the upper left of the input video is the origin (0, 0), and the upper left (Xos, Yos) to the lower right (Xoe, Yoe).
- FIG. 17B shows the coordinate position where the object position is specified on the image classified by distance extracted in FIG. 17A, and the upper left (Xds, The range surrounded by the lower right (Xde, Yde) from Yds) is the detection position.
- FIG. 17C shows the result of converting the coordinates of the object position specified on the video by distance to the coordinates on the input video.
- the upper left of the input video is the origin (0, 0) and the upper left (Xrs, Yrs).
- the range surrounded by the lower right (Xre, Yre) is the coordinates of the final object detection result.
- Xrs, Yrs, Xre, and Yre can be obtained by the following formulas 7 to 10 using the variables shown in FIGS. 17A and 17B.
- the object detection device on the video obtained by converting the video image of the periphery of the vehicle into the characteristic video, and extracting and synthesizing different regions from the characteristic video for each distance based on the distance from the vehicle.
- the object detection position is obtained by performing the first object detection process, and the object detection position is specified after performing the second object detection process on the object candidate position, so that the object detection performance is not sacrificed.
- the speed of the object detection process can be increased, and the distance from the vehicle to the object can also be measured.
- the object detection method of the present invention can be a method including each step realized by the object detection device, and each function realized by the object detection device is also a computer for the object detection program of the present invention. There is no particular limitation as long as it is realized.
- the object detection device converts a video obtained by photographing the periphery of a vehicle into a feature video, and extracts different regions for each distance from the feature video based on the distance from the vehicle.
- the first object detection process is performed to obtain the object candidate position, and the object detection position is specified after performing the second object detection process on the object candidate position, so that the object detection performance is sacrificed.
- it has an excellent effect of speeding up the object detection process and measuring the distance from the vehicle to the object, and is useful as an object detection device for detecting an object around the vehicle.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
Description
以下、本発明の一実施形態に係る物体検出装置について、図面を参照して説明する。
20 映像変換部
30 距離別映像抽出合成部
31 距離別映像抽出部
32 距離別映像拡縮部
33 中心位置合わせ部
34 距離別映像特徴量判定部
35 合成係数調整部
36 距離別映像合成部
40 第一物体検出部
41 合成映像切出拡大部
42 合成映像特徴量判定部
43 合成映像物体検出部
50 物体候補位置指定部
60 第二物体検出部
70 物体位置特定部
Claims (18)
- 車両から撮影した車両周辺の映像を入力映像として、該入力映像から車両付近の物体を検出する物体検出装置であって、
前記入力映像を、該入力映像から画像特徴を抽出した特徴映像へと変換する映像変換部と、
前記車両からの距離に基づいて、前記特徴映像から、距離毎に異なる領域を距離別映像として抽出し、前記距離別映像を用いた合成映像を合成する距離別映像抽出合成部と、
前記距離別映像抽出合成部で合成した前記合成映像を走査して、前記合成映像から検出される物体の前記合成映像上の位置を求める第一の物体検出処理を行う第一物体検出部と、
前記第一物体検出部で検出された物体の前記合成映像上の位置に基づいて、前記距離別映像において前記物体が存在する可能性がある物体候補位置を求める物体候補位置指定部と、
前記物体候補位置指定部で求めた物体候補位置に対して、前記距離別映像において対応する物体位置を特定する第二の物体検出処理を行う第二物体検出部と、
前記第二物体検出部で求めた物体位置に基づいて、前記入力映像における物体位置を特定する物体位置特定部と、
を備えることを特徴とする物体検出装置。 - 前記映像変換部が、前記画像特徴としてエッジ特徴を抽出することを特徴とする請求項1記載の物体検出装置。
- 前記距離別映像抽出合成部が、前記車両からの距離に基づいて、前記特徴映像から抽出する領域の大きさを変更することを特徴とする請求項1に記載の物体検出装置。
- 前記距離別映像抽出合成部が、全ての前記距離別映像の垂直サイズが等しくなるよう、距離毎に抽出した映像を拡縮することを特徴とする請求項1に記載の物体検出装置。
- 前記距離別映像抽出合成部が、前記距離別映像の水平中心軸および垂直中心軸の位置を合わせて合成することを特徴とする請求項1に記載の物体検出装置。
- 前記距離別映像抽出合成部が、前記距離別映像をαブレンディングで合成することを特徴とする請求項1に記載の物体検出装置。
- 前記距離別映像抽出合成部が、前記距離別映像の合成数に応じて、部分的にαブレンディングの合成係数を調整して合成することを特徴とする請求項1に記載の物体検出装置。
- 前記距離別映像抽出合成部が、前記距離別映像に含まれる画像特徴量に応じて、αブレンディングの合成係数を調整して合成することを特徴とする請求項1に記載の物体検出装置。
- 前記第一物体検出部が、前記第一の物体検出処理において、検出対象とする物体の垂直サイズを、前記合成映像の垂直サイズと等しく設定することを特徴とする請求項1に記載の物体検出装置。
- 前記第一物体検出部が、前記合成映像上を水平方向だけに走査して物体検出処理を行うことを特徴とする請求項1に記載の物体検出装置。
- 前記第一物体検出部が、前記合成映像の下端部に接する領域を切り出して、切り出した領域の垂直サイズを前記合成映像の垂直サイズと等しくなるように拡大した拡大合成映像を生成する合成映像切出拡大部を備えることを特徴とする請求項1に記載の物体検出装置。
- 前記第一物体検出部が、前記拡大合成映像に対して物体検出処理を行うことを特徴とする請求項11に記載の物体検出装置。
- 前記第一物体検出部が、前記拡大合成映像上を水平方向だけに走査して物体検出処理を行うことを特徴とする請求項12に記載の物体検出装置。
- 前記第一物体検出部が、物体検出を行う検出判定基準を前記第二物体検出部とは別々に調整可能であって、かつ、前記第二物体検出部よりも物体であると判定しやすくなるように検出基準が調整されていることを特徴とする請求項1に記載の物体検出装置。
- 前記第一物体検出部が、前記合成映像上で画像特徴が存在する部分だけを対象として、第一の物体検出処理を行うことを特徴とする請求項1に記載の物体検出装置。
- 前記物体位置特定部が、前記第二物体検出部で検出された前記距離別映像上の物体位置に基づいて、前記車両から検出された物体までの距離を求めることを特徴とする請求項1に記載の物体検出装置。
- 車両から撮影した車両周辺の映像を入力映像として、該入力映像から車両付近の物体を検出する物体検出方法であって、
前記入力映像を、該入力映像から画像特徴を抽出した特徴映像へと変換する映像変換ステップと、
前記車両からの距離に基づいて、前記特徴映像から、距離毎に異なる領域を距離別映像として抽出し、前記距離別映像を用いた合成映像を合成する距離別映像抽出合成ステップと、
前記距離別映像抽出合成ステップで合成した前記合成映像を走査して、前記合成映像から検出される物体の前記合成映像上の位置を求める第一の物体検出処理を行う第一物体検出ステップと、
前記第一物体検出ステップで検出された物体の前記合成映像上の位置に基づいて、前記距離別映像において前記物体が存在する可能性がある物体候補位置を求める物体候補位置検出ステップと、
前記物体候補位置検出ステップで求めた物体候補位置に対して、前記距離別映像において対応する物体位置を特定する第二の物体検出処理を行う第二物体検出ステップと、
前記第二物体検出ステップで求めた物体位置に基づいて、前記入力映像における物体位置を特定する物体位置特定ステップと、
を含むことを特徴とする物体検出方法。 - 車両から撮影した車両周辺の映像を入力映像として、該入力映像から車両付近の物体を検出する物体検出プログラムであって、
コンピュータに、
前記入力映像を、該入力映像から画像特徴を抽出した特徴映像へと変換する映像変換機能と、
前記車両からの距離に基づいて、前記特徴映像から、距離毎に異なる領域を距離別映像として抽出し合成する距離別映像抽出合成機能と、
前記距離別映像抽出合成機能で合成した合成映像を走査して第一の物体検出処理を行う第一物体検出機能と、
前記第一物体検出機能で物体が検出された前記合成映像上の位置に基づいて、前記距離別映像において物体が存在する可能性がある物体候補位置を求める物体候補位置検出機能と、
前記物体候補位置検出機能で求めた物体候補位置に対して第二の物体検出処理を行い前記距離別映像における物体位置を特定する第二物体検出機能と、
前記第二物体検出機能で求めた物体位置に基づいて、前記入力映像における物体位置を特定する物体位置特定機能と、
を実現させることを特徴とする物体検出プログラム。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013517820A JP5877376B2 (ja) | 2011-06-02 | 2012-04-05 | 物体検出装置、物体検出方法および物体検出プログラム |
CN201280001918.7A CN102985945B (zh) | 2011-06-02 | 2012-04-05 | 物体检测装置、物体检测方法 |
EP12788099.5A EP2717219B1 (en) | 2011-06-02 | 2012-04-05 | Object detection device, object detection method, and object detection program |
US13/672,002 US9152887B2 (en) | 2011-06-02 | 2012-11-08 | Object detection device, object detection method, and object detection program |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011124362 | 2011-06-02 | ||
JP2011-124362 | 2011-06-02 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US13/672,002 Continuation US9152887B2 (en) | 2011-06-02 | 2012-11-08 | Object detection device, object detection method, and object detection program |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012164804A1 true WO2012164804A1 (ja) | 2012-12-06 |
Family
ID=47258682
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2012/002375 WO2012164804A1 (ja) | 2011-06-02 | 2012-04-05 | 物体検出装置、物体検出方法および物体検出プログラム |
Country Status (5)
Country | Link |
---|---|
US (1) | US9152887B2 (ja) |
EP (1) | EP2717219B1 (ja) |
JP (1) | JP5877376B2 (ja) |
CN (1) | CN102985945B (ja) |
WO (1) | WO2012164804A1 (ja) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20150102525A (ko) * | 2014-02-28 | 2015-09-07 | 경북대학교 산학협력단 | 동등-높이 정합 영상을 이용하여 물체를 검출하기 위한 영상 처리 장치 및 방법, 그리고 그를 이용한 차량 운전 보조 시스템 |
KR101593484B1 (ko) * | 2014-07-10 | 2016-02-15 | 경북대학교 산학협력단 | 동등-높이 주변영역 정합 영상을 이용하여 측면에서 접근하는 일부분만 보이는 물체를 검출하기 위한 영상 처리 장치 및 방법, 그리고 그를 이용한 차량 운전 보조 시스템 |
CN108491795A (zh) * | 2018-03-22 | 2018-09-04 | 北京航空航天大学 | 轨道交通场景的行人检测方法与装置 |
KR20200091331A (ko) * | 2019-01-22 | 2020-07-30 | 주식회사 스트라드비젼 | 다중 카메라 혹은 서라운드 뷰 모니터링에 이용되기 위해, 타겟 객체 통합 네트워크 및 타겟 영역 예측 네트워크를 이용하여 핵심성과지표와 같은 사용자 요구 사항에 적응 가능한 cnn 기반 객체 검출기를 학습하는 방법 및 학습 장치, 그리고 이를 이용한 테스팅 방법 및 테스팅 장치 |
JP2021174147A (ja) * | 2020-04-22 | 2021-11-01 | 富士通クライアントコンピューティング株式会社 | 情報処理装置、情報処理システムおよびプログラム |
WO2024069778A1 (ja) * | 2022-09-28 | 2024-04-04 | 株式会社日立国際電気 | 物体検知システム、カメラ、及び物体検知方法 |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014103433A1 (ja) * | 2012-12-25 | 2014-07-03 | 本田技研工業株式会社 | 車両周辺監視装置 |
JP6471528B2 (ja) * | 2014-02-24 | 2019-02-20 | 株式会社リコー | 物体認識装置、物体認識方法 |
KR102199094B1 (ko) * | 2014-05-26 | 2021-01-07 | 에스케이텔레콤 주식회사 | 관심객체 검출을 위한 관심영역 학습장치 및 방법 |
RU2714091C1 (ru) * | 2016-06-27 | 2020-02-11 | Ниссан Мотор Ко., Лтд. | Способ отслеживания объектов и устройство отслеживания объектов |
JP2018136803A (ja) * | 2017-02-23 | 2018-08-30 | 株式会社日立製作所 | 画像認識システム |
KR101958275B1 (ko) * | 2017-07-07 | 2019-03-14 | 한국항공우주연구원 | 영상 패치 정규화 방법 및 시스템 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001052171A (ja) * | 1999-08-06 | 2001-02-23 | Nissan Motor Co Ltd | 周囲環境認識装置 |
JP2007265390A (ja) | 2006-02-28 | 2007-10-11 | Sanyo Electric Co Ltd | 物体検出装置 |
JP2007272555A (ja) * | 2006-03-31 | 2007-10-18 | Victor Co Of Japan Ltd | 画像処理装置 |
JP2007316790A (ja) * | 2006-05-24 | 2007-12-06 | Nissan Motor Co Ltd | 歩行者検出装置および歩行者検出方法 |
JP2011055366A (ja) * | 2009-09-03 | 2011-03-17 | Panasonic Corp | 画像処理装置及び画像処理方法 |
Family Cites Families (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7243945B2 (en) * | 1992-05-05 | 2007-07-17 | Automotive Technologies International, Inc. | Weight measuring systems and methods for vehicles |
US7415126B2 (en) * | 1992-05-05 | 2008-08-19 | Automotive Technologies International Inc. | Occupant sensing system |
US7477758B2 (en) * | 1992-05-05 | 2009-01-13 | Automotive Technologies International, Inc. | System and method for detecting objects in vehicular compartments |
WO1997016807A1 (en) * | 1995-10-31 | 1997-05-09 | Sarnoff Corporation | Method and apparatus for image-based object detection and tracking |
US6757009B1 (en) * | 1997-06-11 | 2004-06-29 | Eaton Corporation | Apparatus for detecting the presence of an occupant in a motor vehicle |
JP3298851B2 (ja) * | 1999-08-18 | 2002-07-08 | 松下電器産業株式会社 | 多機能車載カメラシステムと多機能車載カメラの画像表示方法 |
JP3599639B2 (ja) * | 2000-05-26 | 2004-12-08 | 松下電器産業株式会社 | 画像処理装置 |
JP2002359839A (ja) * | 2001-03-29 | 2002-12-13 | Matsushita Electric Ind Co Ltd | リアビューカメラの画像表示方法及びその装置 |
JP2003016429A (ja) * | 2001-06-28 | 2003-01-17 | Honda Motor Co Ltd | 車両周辺監視装置 |
US20030137586A1 (en) * | 2002-01-22 | 2003-07-24 | Infinite Innovations, Inc. | Vehicle video switching system and method |
EP1398601A3 (en) * | 2002-09-13 | 2014-05-07 | Canon Kabushiki Kaisha | Head up display for navigation purposes in a vehicle |
US7782374B2 (en) * | 2005-03-03 | 2010-08-24 | Nissan Motor Co., Ltd. | Processor and processing method for generating a panoramic image for a vehicle |
EP1901225A1 (en) * | 2005-05-10 | 2008-03-19 | Olympus Corporation | Image processing device, image processing method, and image processing program |
US8885045B2 (en) * | 2005-08-02 | 2014-11-11 | Nissan Motor Co., Ltd. | Device and method for monitoring vehicle surroundings |
EP2000889B1 (en) * | 2006-03-15 | 2018-06-27 | Omron Corporation | Monitor and monitoring method, controller and control method, and program |
EP2168079B1 (en) * | 2007-01-23 | 2015-01-14 | Valeo Schalter und Sensoren GmbH | Method and system for universal lane boundary detection |
JP4863922B2 (ja) * | 2007-04-18 | 2012-01-25 | 三洋電機株式会社 | 運転支援システム並びに車両 |
EP2674323B1 (en) * | 2007-04-30 | 2018-07-11 | Mobileye Vision Technologies Ltd. | Rear obstruction detection |
JP4595976B2 (ja) * | 2007-08-28 | 2010-12-08 | 株式会社デンソー | 映像処理装置及びカメラ |
JP4450036B2 (ja) * | 2007-09-10 | 2010-04-14 | トヨタ自動車株式会社 | 複合画像生成装置、及びプログラム |
US8988525B2 (en) * | 2009-08-27 | 2015-03-24 | Robert Bosch Gmbh | System and method for providing guidance information to a driver of a vehicle |
WO2011028686A1 (en) * | 2009-09-01 | 2011-03-10 | Magna Mirrors Of America, Inc. | Imaging and display system for vehicle |
JP4689758B1 (ja) * | 2010-04-22 | 2011-05-25 | 株式会社市川ソフトラボラトリー | 画像一致点検出装置、画像一致点検出方法および記録媒体 |
US10089537B2 (en) * | 2012-05-18 | 2018-10-02 | Magna Electronics Inc. | Vehicle vision system with front and rear camera integration |
-
2012
- 2012-04-05 WO PCT/JP2012/002375 patent/WO2012164804A1/ja active Application Filing
- 2012-04-05 CN CN201280001918.7A patent/CN102985945B/zh active Active
- 2012-04-05 EP EP12788099.5A patent/EP2717219B1/en active Active
- 2012-04-05 JP JP2013517820A patent/JP5877376B2/ja active Active
- 2012-11-08 US US13/672,002 patent/US9152887B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001052171A (ja) * | 1999-08-06 | 2001-02-23 | Nissan Motor Co Ltd | 周囲環境認識装置 |
JP2007265390A (ja) | 2006-02-28 | 2007-10-11 | Sanyo Electric Co Ltd | 物体検出装置 |
JP2007272555A (ja) * | 2006-03-31 | 2007-10-18 | Victor Co Of Japan Ltd | 画像処理装置 |
JP2007316790A (ja) * | 2006-05-24 | 2007-12-06 | Nissan Motor Co Ltd | 歩行者検出装置および歩行者検出方法 |
JP2011055366A (ja) * | 2009-09-03 | 2011-03-17 | Panasonic Corp | 画像処理装置及び画像処理方法 |
Non-Patent Citations (1)
Title |
---|
See also references of EP2717219A4 |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20150102525A (ko) * | 2014-02-28 | 2015-09-07 | 경북대학교 산학협력단 | 동등-높이 정합 영상을 이용하여 물체를 검출하기 위한 영상 처리 장치 및 방법, 그리고 그를 이용한 차량 운전 보조 시스템 |
KR101593483B1 (ko) * | 2014-02-28 | 2016-02-12 | 경북대학교 산학협력단 | 동등-높이 정합 영상을 이용하여 물체를 검출하기 위한 영상 처리 장치 및 방법, 그리고 그를 이용한 차량 운전 보조 시스템 |
US9646363B2 (en) | 2014-02-28 | 2017-05-09 | Kyungpook National University Industry-Academic Cooperation Foundation | Image processing apparatus and method for detecting object using equi-height mosaicking image, and vehicle operation assisting system employing same |
KR101593484B1 (ko) * | 2014-07-10 | 2016-02-15 | 경북대학교 산학협력단 | 동등-높이 주변영역 정합 영상을 이용하여 측면에서 접근하는 일부분만 보이는 물체를 검출하기 위한 영상 처리 장치 및 방법, 그리고 그를 이용한 차량 운전 보조 시스템 |
US9569685B2 (en) | 2014-07-10 | 2017-02-14 | Kyungpook National University Industry-Academic Cooperation Foundation | Image processing apparatus and method for detecting partially visible object approaching from side using equi-height peripheral mosaicking image, and driving assistance system employing the same |
CN108491795A (zh) * | 2018-03-22 | 2018-09-04 | 北京航空航天大学 | 轨道交通场景的行人检测方法与装置 |
CN108491795B (zh) * | 2018-03-22 | 2022-05-13 | 北京航空航天大学 | 轨道交通场景的行人检测方法与装置 |
KR20200091331A (ko) * | 2019-01-22 | 2020-07-30 | 주식회사 스트라드비젼 | 다중 카메라 혹은 서라운드 뷰 모니터링에 이용되기 위해, 타겟 객체 통합 네트워크 및 타겟 영역 예측 네트워크를 이용하여 핵심성과지표와 같은 사용자 요구 사항에 적응 가능한 cnn 기반 객체 검출기를 학습하는 방법 및 학습 장치, 그리고 이를 이용한 테스팅 방법 및 테스팅 장치 |
KR102328731B1 (ko) | 2019-01-22 | 2021-11-22 | 주식회사 스트라드비젼 | 다중 카메라 혹은 서라운드 뷰 모니터링에 이용되기 위해, 타겟 객체 통합 네트워크 및 타겟 영역 예측 네트워크를 이용하여 핵심성과지표와 같은 사용자 요구 사항에 적응 가능한 cnn 기반 객체 검출기를 학습하는 방법 및 학습 장치, 그리고 이를 이용한 테스팅 방법 및 테스팅 장치 |
JP2021174147A (ja) * | 2020-04-22 | 2021-11-01 | 富士通クライアントコンピューティング株式会社 | 情報処理装置、情報処理システムおよびプログラム |
WO2024069778A1 (ja) * | 2022-09-28 | 2024-04-04 | 株式会社日立国際電気 | 物体検知システム、カメラ、及び物体検知方法 |
Also Published As
Publication number | Publication date |
---|---|
US9152887B2 (en) | 2015-10-06 |
EP2717219B1 (en) | 2018-05-30 |
JPWO2012164804A1 (ja) | 2014-07-31 |
CN102985945B (zh) | 2016-09-07 |
US20130070096A1 (en) | 2013-03-21 |
EP2717219A4 (en) | 2016-05-25 |
JP5877376B2 (ja) | 2016-03-08 |
CN102985945A (zh) | 2013-03-20 |
EP2717219A1 (en) | 2014-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5877376B2 (ja) | 物体検出装置、物体検出方法および物体検出プログラム | |
JP4930046B2 (ja) | 路面判別方法および路面判別装置 | |
JP4171501B2 (ja) | 車両の周辺監視装置 | |
JP5959073B2 (ja) | 検出装置、検出方法、及び、プログラム | |
JP4872769B2 (ja) | 路面判別装置および路面判別方法 | |
JP5401257B2 (ja) | 遠赤外線歩行者検知装置 | |
EP2237988A2 (en) | Object detection and recognition system | |
EP3115966A1 (en) | Object detection device, object detection method, and computer program | |
JP2018025906A (ja) | 画像処理装置、撮像装置、移動体機器制御システム、画像処理方法、及びプログラム | |
JP2018147393A (ja) | 標識認識システム | |
JP2012252501A (ja) | 走行路認識装置及び走行路認識用プログラム | |
JP2017207874A (ja) | 画像処理装置、撮像装置、移動体機器制御システム、画像処理方法、及びプログラム | |
JP6711128B2 (ja) | 画像処理装置、撮像装置、移動体機器制御システム、画像処理方法、及びプログラム | |
JP5062091B2 (ja) | 移動体識別装置、コンピュータプログラム及び光軸方向特定方法 | |
US10789727B2 (en) | Information processing apparatus and non-transitory recording medium storing thereon a computer program | |
JP2012185555A (ja) | 物体検出装置及び方法、並びにプログラム | |
JP2011033594A (ja) | 車両用距離算出装置 | |
JP2006003994A (ja) | 道路標識認識装置 | |
JP6802999B2 (ja) | 区画線検出システム | |
EP3287948A1 (en) | Image processing apparatus, image capturing apparatus, moving body apparatus control system, image processing method, and program | |
JP2002150302A (ja) | 路面認識装置 | |
CN111133439B (zh) | 全景监视系统 | |
JP2010239448A (ja) | 道路標識認識装置 | |
KR20150111611A (ko) | 차량 후보 검출 장치 및 그 방법 | |
JP2004310282A (ja) | 車両検出装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201280001918.7 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2012788099 Country of ref document: EP |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 12788099 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2013517820 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |