KR20190098856A - Method and apparatus for blind spot detection - Google Patents

Abstract

Disclosed are a method for detecting a rear/side vehicle and an apparatus thereof. The method for detecting a rear/side vehicle comprises the steps of: (a) obtaining a rear image of a vehicle; (b) applying a filter to the rear image of the vehicle to detect a wheel area of the vehicle; and (c) determining whether the rear/side vehicle is in proximity in accordance with the wheel area detection result.

Description

Lateral rear vehicle detection method and apparatus therefor {Method and apparatus for blind spot detection}

The present invention relates to a method for detecting a rear and rear vehicle and an apparatus thereof.

BSD stands for Blind Spot Detection, and it is difficult for a driver to observe the near-back rectangular area created by the C-pillar of an automobile. For this reason, a notification is generated in order to prevent a collision caused by a blind spot that is difficult to recognize by detecting a presence or absence of an object using a sensor such as an ultrasonic wave in the near-back rectangular blind spot.

However, in the case of using a sensor such as a conventional ultrasonic wave, there is a problem that causes a driver's fatigue with frequent notifications due to a high false recognition rate.

Disclosure of Invention The present invention provides a method and apparatus for detecting a rear-side vehicle that can accurately detect and inform whether a rear-side vehicle is close by image recognition.

According to an aspect of the present invention, there is provided a side rear vehicle detection method capable of accurately detecting and notifying whether the side rear vehicle is close by image recognition.

According to one embodiment of the invention, (a) obtaining a vehicle rear image; (b) detecting a wheel area of the vehicle by applying a filter to the rear image of the vehicle; And (c) determining whether the rear rear vehicle is in proximity according to the wheel region detection result.

Before the step (c), detecting whether the object invaded the region of interest through the analysis of the marine rear image; Classifying whether an object in a region of interest is a vehicle by analyzing the rear image of the vehicle through an artificial neural network; And detecting an object around a wheel in the ROI, wherein step (c) further includes using the saliency detection result, the object classification result, and the wheel peripheral object detection result. It may be determined whether the vehicle is close.

The step (c) may be used to determine the proximity of the side rear vehicle by applying different weights according to the collision detection result, the wheel detection result, and the wheel surrounding object detection result according to the weather or the illumination.

In the step (c), if at least two or more of the saliency detection result, the wheel region detection result, the object classification result, and the wheel surrounding object detection result are detected in a plurality of frames, it may be determined to be close to the rear-rear vehicle. have.

The filter may be a dark-light-dark filter.

Prior to the step (b), the method may further include setting a region of interest for the rear image of the vehicle, wherein the size of the region of interest may be set differently according to the speed of the vehicle.

The setting of the ROI may include detecting a road lane of the vehicle in the image of the rear of the vehicle and deriving a vanishing point using the road lane; And extending the vanishing point horizontally, excluding a predetermined lower portion of the rear image of the vehicle, and setting a region of interest to a closed shape including a part of left and right roads of the vehicle excluding the road of the vehicle. .

According to another aspect of the present invention, there is provided a side rear vehicle detection apparatus capable of accurately detecting and notifying whether the side rear vehicle is close by image recognition.

According to an embodiment of the present invention, the image acquisition unit for obtaining a vehicle rear image; A wheel detector detecting a wheel area of the vehicle by applying a filter to the rear image of the vehicle; And an analyzer configured to determine whether the vehicle is in proximity to the rear rear vehicle according to the wheel area detection result.

A saliency detection unit that detects whether an object invades a region of interest through a saliency analysis of the rear image of the vehicle; A neural network classifier classifying whether the object in the ROI is a vehicle by analyzing the rear image of the vehicle through an artificial neural network; And a wheel peripheral detector configured to detect a wheel peripheral object in the ROI, wherein the analyzer is further configured to approach the rear-rear vehicle by further using the saliency detection result, the object classification result, and the wheel peripheral object detection result. It can be determined.

By providing a method and apparatus for detecting a rear-side vehicle according to an embodiment of the present invention, it is possible to accurately detect and notify the proximity of the rear-side vehicle through image recognition.

1 is a flow chart showing a side rear vehicle detection method according to an embodiment of the present invention.
2 and 3 are diagrams for explaining a region of interest according to an embodiment of the present invention.
4 is a diagram illustrating a detection result of a marine interference according to an embodiment of the present invention.
5 is a view illustrating a wheel detection according to an embodiment of the present invention.
Figure 6 is a block diagram showing the internal configuration of the side rear vehicle detection apparatus according to an embodiment of the present invention.
7 is a view for explaining the rear and rear vehicle detection according to an embodiment of the present invention.

As used herein, the singular forms "a", "an" and "the" include plural forms unless the context clearly indicates otherwise. In this specification, terms such as “consisting of” or “comprising” should not be construed as necessarily including all of the various components or steps described in the specification, and some of the components or some steps It should be construed that it may not be included or may further include additional components or steps. In addition, the terms "... unit", "module", etc. described in the specification mean a unit for processing at least one function or operation, which may be implemented in hardware or software or a combination of hardware and software. .

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a method of detecting a rear and rear vehicle according to an exemplary embodiment of the present invention, and FIGS. 2 and 3 are views illustrating an ROI according to an exemplary embodiment of the present invention, and FIG. FIG. 5 is a view illustrating a detection result of a mariner according to an embodiment of the present invention, and FIG. 5 is a view illustrating a wheel detection according to an embodiment of the present invention.

In operation 110, the apparatus for detecting rear and rear vehicles 100 acquires a rear image of the vehicle.

For example, the lateral rear vehicle detecting apparatus 100 may acquire a vehicle rear image through the vehicle rear camera. In this case, the rear image of the vehicle may be acquired at least about seven times per second in order to not miss a vehicle approaching at high speed.

In operation 115, the apparatus for detecting rear and rear vehicles 100 sets a region of interest ROI in the acquired vehicle rear image.

For example, the lateral rear vehicle detecting apparatus 100 may obtain the horizon height and the rear trunk height from the vehicle rear image and then set the ROI using the same. According to an embodiment of the present disclosure, the apparatus for detecting the rear and rear vehicles 100 detects both lanes of the road in the rear area of the vehicle, connects the detected lanes with one straight line, and then derives the point where the two straight lines meet as a vanishing point. Calculated vanishing point can be calculated as the height of the horizon. In deriving the height of the trunk area of the vehicle from the vehicle rear area, the side rear monitoring apparatus 100 may set the lower 1/5 area as the default area when the trunk line of the vehicle is not visible. For this reason, as shown in FIG. 2, the side rear vehicle detecting apparatus 100 excludes a certain region of the lower end from the image of the rear side of the vehicle to minimize the influence of the vehicle, and then horizontally connects the vanishing point (ie, the horizontal line). A line extending horizontally) may be set as a region of interest.

The apparatus for detecting the rear side vehicle 100 according to an embodiment of the present invention has a main purpose of informing a driver of a vehicle approaching a blind spot in the rear side. Therefore, the apparatus for detecting rear-rear vehicle 100 may set an area corresponding to both side lanes of the lane in which the vehicle is driving, as the monitoring area, without setting the entire ROI as the monitoring area.

Accordingly, as shown in FIG. 3, the lateral rear vehicle detecting apparatus 100 determines the scale and shift of the mask map with the shape of a human lung to determine the final ROI. Can be set to

In addition, the side rear vehicle detection apparatus 100 may set the ROI differently according to the speed of the vehicle. On highways, roads are usually wider than regular roads. Therefore, when the vehicle is a high speed of the reference value or more, the side rear vehicle detection apparatus 100 may set the size of the ROI wide in the horizontal direction. On the other hand, when the vehicle is less than the reference value, the rear and rear vehicle detection apparatus 100 may set the size of the ROI to be narrow in the horizontal direction.

In operation 120, the apparatus for detecting rear-rear vehicle 100 detects whether an object is invaded by performing a marine area detection on an image of the rear of the vehicle.

For example, the lateral rear vehicle detecting apparatus 100 performs the saliency region detection on the rear image of the vehicle, and then masks the saliency region detection result on the region of interest (for example, the mask map) to the left and right sides. It can detect whether an object is invaded.

For another example, the side rear vehicle detection apparatus 100 may detect whether an object is invaded by performing a marine area detection on the ROI. Fig. 4 shows the result of the salient detection. Since the salient area detection is obvious to those skilled in the art, a separate description thereof will be omitted.

In operation 125, the apparatus for detecting rear-rear vehicle 100 applies a filter to a region of interest to detect a region estimated by a wheel.

When the side and rear surfaces of the vehicle are photographed by a wide-angle monocular rear camera, tires, wheels, and wheelbase shade regions of the vehicle are commonly displayed regardless of the vehicle model (see FIG. 5). Accordingly, the lateral rear vehicle detecting apparatus 100 may estimate the elliptical wheelbase region of the vehicle through image search and recognition in the ROI (ie, the mask region) and perform image recognition detection thereof. To this end, the rear and rear vehicle detection apparatus 100 according to an embodiment of the present invention detects an area estimated as a wheel by using a dark-light-dark filter. In other words, the wheels of a car appear as a series of "tire-wheel-tire" when it follows a horizontal or vertical center line. Therefore, the rear-rear vehicle detecting apparatus 100 searches the entire region of interest (ie, the mask region) and applies the dark-light-dark filter to detect the region estimated by the automobile wheel.

In operation 130, the apparatus 100 for detecting rear and rear vehicles classifies whether an object detected in the ROI is a vehicle using an artificial neural network.

For example, the side and rear vehicle detection apparatus 100 detects the edge twice in an oblique direction (for example, 45 degrees, 135 degrees) so as to accurately detect lane or guardrail information of the ROI. Can be entered as a feature value in the neural network. In addition, the side rear vehicle detection apparatus 100 may input the pixel color information near the edge as a feature value to accurately classify the object corresponding to the current edge.

In operation 132, the apparatus for detecting rear and rear vehicles 100 detects objects around the wheel in the ROI. Here, the object around the wheel may include, for example, at least one of a vehicle headlight, a fender, and a wheel.

For example, the lateral rear vehicle detecting apparatus 100 may detect a candidate object using a Haar-like feature algorithm in the ROI. Haar-like Feature As a result, guardrails, road signs, crosswalks, and sidewalk blocks are likely to be misidentified. Therefore, a negative object may be collected based on a random image that may appear in a driving environment, and then may have a DB. Accordingly, the lateral rear vehicle detecting apparatus 100 may exclude all objects included in the negative object after detecting candidate objects using the Haar-like Feature algorithm in the ROI.

In applying Haar-like Feature, cascade stage can be applied in 7 stages and limited to 20 features in each stage. In addition, the subwindow size may be limited to 32 or 40.

The apparatus for detecting rear-rear vehicle 100 may finally detect objects around wheels by applying HOG to candidate objects detected as a result of a Haar-like Feature. Here, the HOG may be normalized to 32 × 32 to extract objects around the wheel.

In operation 135, the apparatus for detecting rear-rear vehicle 100 determines whether the vehicle is close to the rear-rear vehicle by using the collision detection result, the wheel area detection result, the neural network classification result, and the wheel surrounding object detection result.

In this case, the side rear vehicle detection apparatus 100 may determine whether the side rear vehicle is close by applying different weights to the marine detection result and the wheel area detection result according to the weather or the illumination. For example, a sunny day may be affected by the detection of saliency by roadsides or shadows on the buildings. On the other hand, in sunny weather, the wheel reflects well, and wheel detection results may be better. Accordingly, the side rear vehicle detection apparatus 100 may set a low weight according to the mariner detection result on a good day and set a high weight on the wheel detection result.

In another example, when the weather is cloudy or dark, the result of applying the dark-light-dark filter may not be accurate. Therefore, on a cloudy or dark day, the weight may be set high in the sacrificial detection result and the weight may be set low in the wheel detection result. On the other hand, the weight of the object search results around the wheel using the shape information can be set high.

The side and rear vehicle detection apparatus 100 may assign a score according to the collision detection result, the wheel area detection result, the neural network classification result, and the wheel surrounding object detection result for each frame, and the reference point based on the cumulative score of four frames. In case of abnormality, it may be determined that the vehicle exists.

In order to facilitate understanding and explanation, a description will be given of a method of determining whether a vehicle is close to the rear and rear using a cumulative score after describing the score criteria for each frame.

For example, the side rear vehicle detection apparatus 100 scores a frame for a frame when the object detection is successful, the wheel detection is successful, the wheel detection is successful, and the neural network classification is classified as a vehicle or asphalt. Can be set to points.

In addition, the side rear vehicle detecting apparatus 100 may set the score for the frame to two points when it is determined that the object around the wheel is a successful detection and the neural network classification result indicates that the object is a vehicle, asphalt or road sign.

In other cases, the score of the frame may be set to -2.

Based on the above-described example, it will be assumed that each frame score and cumulative score according to the collision detection result, the wheel region detection result, the neural network classification result, and the wheel surrounding object detection result are the same as in FIG. 7.

As shown in FIG. 4, when the object detection is successful in the t3 frame, the object detection is successful, the wheel detection is successful, the wheel detection is successful, and the neural network classification result is classified as a vehicle or asphalt. It is set to, it is assumed that the cumulative score is 1. In this case, the rear and rear vehicle detecting apparatus 100 may not generate a warning.

In case of the same result as the t-3 frame in the t-2 frame, the score of the t-2 frame is 1 point, but the cumulative score is 2 points. Therefore, the rear-rear vehicle detecting apparatus 100 generates a warning (notification) according to the proximity of the rear-rear vehicle according to the collision detection result, the wheel area detection result, the neural network classification result, and the wheel surrounding object detection result in the t-2 frame. Done.

If the result is the same as the t-3 and t-2 frames even in the t-1 frame, the score of the t-1 frame is one point but the cumulative score is three points. In the frame, a warning (notification) is generated according to the proximity of the rear and rear vehicles according to the result of the collision detection, the wheel region, the neural network classification, and the object around the wheel.

In the case of the t frame, when the object detection is successful, the wheel detection result is successful, the wheel detection result is detected, the object around the wheel is detected, the object around the wheel is detected, and the neural network classification is determined as a vehicle, asphalt, or road sign. The score of the t frame may be set to three points. In this case, the side rear vehicle detection apparatus 100 may provide a warning (notification) according to the proximity of the rear rear vehicle according to the collision detection result, the wheel area detection result, the neural network classification result, and the wheel surrounding object detection result for the t frame. Will occur.

As such, the apparatus for detecting the rear-rear vehicle 100 may generate a warning (notification) according to the proximity of the rear-rear vehicle using the cumulative result of the previous four frames.

In addition, when the collision detection result object detection, the wheel detection result wheel detection is successful, but the neural network classification result is classified as not the vehicle, the side rear vehicle detection device 100 is maintained in the same state for more than two frames It can be determined that the vehicle is close to the rear and rear.

In another example, when the object detection is successful, the wheel detection is unsuccessful, and the wheel detection is unsuccessful, the neural network classification result is classified as a vehicle. If it is determined that the vehicle is close to the rear and rear.

As described above, the at least two rear vehicle detecting apparatuses 100 may determine the proximity of the rear rear vehicle when at least two or more results of the collision detection result, the wheel detection result, and the neural network classification result are detected.

6 is a block diagram showing an internal configuration of a side rear vehicle detection apparatus according to an embodiment of the present invention.

Referring to FIG. 6, the apparatus for detecting rear and rear vehicles 100 according to an exemplary embodiment of the present invention may include an image acquisition unit 610, a preprocessor 615, a saliency detector 620, a wheel detector 625, The neural network classifier 630, the analyzer 635, the notification generator 640, the memory 645, and the processor 650 are configured to be included.

The image acquisition unit 610 is a means for acquiring a rear image of the vehicle. For example, the image acquirer 610 may acquire a rear image of the vehicle through a rear camera located at the rear of the vehicle. In this case, the image acquisition unit 610 may be acquired at least about seven times per second to detect the high speed proximity vehicle without missing it.

The preprocessor 615 is a means for setting a region of interest in the vehicle rear image. Since this is the same as already described above, redundant description will be omitted.

The salient detector 620 is a means for detecting whether an object is invaded in the region of interest by performing a saliency analysis of the vehicle rear image or the region of interest.

The wheel detector 625 is a means for detecting a wheel region by applying a dark-light-dark filter to the region of interest.

The neural network classifier 630 is a means for classifying whether an object detected in the ROI is a vehicle using an artificial neural network.

The wheel perimeter detector 632 is a means for detecting the wheel perimeter in the region of interest. For example, the wheel peripheral detector 632 may extract a candidate object by applying a Haar-like feature in the ROI and finally detect the object around the wheel by applying a HOG. The analyzing unit 635 is a means for determining the proximity of the rear and rear vehicles using the collision detection result, the wheel detection result, the neural network classification result, and the wheel surrounding object detection result.

For example, if the collision detection result object detection is successful, the wheel detection result wheel detection is successful, but the neural network classification result is classified as not the vehicle, the analysis unit 635 side It can be determined that there is a possibility that the vehicle exists in the rear. Since this is the same as described with reference to FIG. 7, redundant descriptions will be omitted.

The notification generator 640 generates a notification when it is determined that the analysis unit 635 determines the proximity of the rear and rear vehicle.

The memory 645 is a means for storing various algorithms, various data derived in this process, etc. necessary to perform the side rear vehicle detection method according to an embodiment of the present invention.

The processor 650 may include internal components (eg, the image acquisition unit 610, the preprocessing unit 615, and the saliency detection unit 620) of the side and rear vehicle detection apparatus 100 according to an exemplary embodiment. ), A wheel detector 625, a neural network classifier 630, an analyzer 635, a notification generator 640, a memory 645, and the like.

The above-described lane prohibition lane warning method according to the present invention can be embodied as computer readable codes on a computer readable recording medium. Computer-readable recording media include all kinds of recording media having data stored thereon that can be decrypted by a computer system. For example, there may be a read only memory (ROM), a random access memory (RAM), a magnetic tape, a magnetic disk, a flash memory, an optical data storage device, and the like. The computer readable recording medium can also be distributed over computer systems connected over a computer network, stored and executed as readable code in a distributed fashion.

100: rear and rear vehicle detection device
610: image acquisition unit
615: preprocessing unit
620: the salient detection unit
625: wheel detection unit
630: neural network classifier
632: wheel peripheral detection unit
635: analysis unit
640: notification generator
645: memory
650: processor

Claims (10)

(a) acquiring a rear image of the vehicle;
(b) detecting a wheel area of the vehicle by applying a filter to the rear image of the vehicle; And
(c) determining whether the rear rear vehicle is in proximity according to the wheel region detection result.
According to claim 1,
Before step (c),
Detecting whether an object is invaded in a region of interest by using the analysis of the vehicle rear image;
Classifying whether an object in a region of interest is a vehicle by analyzing the rear image of the vehicle through an artificial neural network; And
Detecting an object around a wheel in the region of interest;
In step (c),
And determining whether the side rear vehicle is close by further using the collision detection result, the object classification result, and the wheel surrounding object detection result.
The method of claim 2,
In step (c),
The rear and rear vehicle detection method according to claim 1, wherein the weight is applied differently according to the collision detection result, the wheel detection result and the wheel surrounding object detection result according to the weather or the illumination.
The method of claim 2,
In step (c),
If the at least two or more results of the collision detection result, the wheel area detection result, the object classification result and the wheel surrounding object detection result in a plurality of frames is successful detection, the rear and rear vehicle is determined to be close to the rear and rear vehicle Detection method.
According to claim 1,
Said filter is a dark-light-dark filter.
According to claim 1,
Before step (b),
The method may further include setting a region of interest with respect to the rear image of the vehicle.
The size of the ROI is differently set according to the speed of the vehicle.
The method of claim 6,
The setting of the ROI may include:
Detecting road traffic lanes of the vehicle from the vehicle rear image and deriving a vanishing point using the road traffic lanes; And
And extending the vanishing point horizontally, excluding a predetermined lower portion of the rear image of the vehicle, and setting a region of interest to a closed shape including a portion of left and right roads of the vehicle excluding the road of the vehicle. Lateral rear vehicle detection method.
A computer-readable recording medium product having recorded thereon program codes for carrying out a method of detecting a rear vehicle, comprising:
(a) acquiring a rear image of the vehicle;
(b) detecting a wheel area of the vehicle by applying a filter to the rear image of the vehicle; And
and (c) determining whether the vehicle is in proximity to the rear and rear vehicle according to the wheel area detection result.
An image acquisition unit for acquiring a rear image of the vehicle;
A wheel detector detecting a wheel area of the vehicle by applying a filter to the rear image of the vehicle; And
And a rear side rear vehicle detecting apparatus including a rear side rear side vehicle determining unit according to the result of the wheel area detection.
The method of claim 9,
A saliency detection unit that detects whether an object invades a region of interest through a saliency analysis of the rear image of the vehicle;
A neural network classifier classifying whether the object in the ROI is a vehicle by analyzing the rear image of the vehicle through an artificial neural network; And
Further comprising a wheel peripheral detector for detecting the object around the wheel in the region of interest,
The analysis unit,
And determining whether the vehicle is in proximity to the rear-rear vehicle by further using the mariner detection result, the object classification result, and the wheel object detection result.

Images