JP2013246794A - Object change detection device and object change detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object change detection device capable of quickly and accurately detecting changes in objects in a periphery of a vehicle, and an object change detection method.SOLUTION: The object change detection device detects changes in an object from an image, acquires a feature quantity distribution representing shape information of the detected object, acquires shape change information of the object representing time changes in the shape information in a prescribed time on the basis of the shape information represented by the acquired feature quantity distribution, sets a threshold for change detection for use in detecting changes in the object on the basis of the acquired shape change information, and detects the changes in the object using the acquired shape change information and the set threshold for the change detection.

Description

  The present invention relates to an object change detection device and an object change detection method.

  Conventionally, techniques for detecting an object around a vehicle have been developed. For example, in Patent Document 1, the time series change of the position and movement speed of the pedestrian existing in front of the host vehicle and the peripheral information are acquired, the time series change of the acquired position and movement speed, and the pedestrian By comparing the position and movement speed pattern of time-series changes when jumping onto the roadway, and comparing the acquired surrounding information with the previously obtained surrounding information when the pedestrian jumps onto the roadway, There is disclosed a pedestrian pop-up prediction device that predicts whether or not a pedestrian will jump out on a roadway on which the vehicle is traveling.

JP 2010-102437 A

  However, the conventional technique has a problem that a change in an object around the vehicle cannot be detected quickly and accurately. For example, in the pedestrian pop-out prediction device described in Patent Document 1, since only the pedestrian's position change and speed change are observed, the pedestrian's jump-out cannot be predicted until the position and speed change can be confirmed. There is a problem that the prediction of the pedestrian's popping out becomes slower than the case of predicting from the posture (that is, the shape of the object).

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an object change detection device and an object change detection method capable of detecting a change in an object around a vehicle quickly and accurately. .

  An object change detection apparatus according to the present invention includes an object detection unit that detects an object from an image, and a shape information acquisition unit that acquires a feature amount distribution representing shape information of the object detected by the object detection unit. Shape change information acquisition for acquiring shape change information of the object indicating a time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired by the shape information acquisition means Based on the shape change information acquired by the shape change information acquisition means, threshold setting means for setting a change detection threshold used for change detection of the object, and the shape change information acquisition means acquired by the shape change information acquisition means Object change detection means for detecting a change in the object using shape change information and the change detection threshold set by the threshold setting means; Characterized in that it obtain.

  In the object change detection device described above, the object change detection means stores the shape change information storage means for storing the shape change information acquired by the shape change information acquisition means and the shape change information storage means. A shape change averaging means for averaging a plurality of the shape change information and obtaining averaged shape change information, and calculating the similarity using the averaged shape change information acquired by the shape change averaging means, It is preferable that the apparatus further includes a change detection unit that detects a change in the object when the similarity is smaller than the change detection threshold set by the threshold setting unit.

  In the object change detection device described above, the threshold setting unit calculates an absolute value of the shape change information acquired by the shape change information acquisition unit or a variance value for a past predetermined time, and the absolute value or the variance When the value is smaller than the predetermined threshold, the change detection threshold is set to a value that decreases the detection sensitivity, and when the absolute value or the variance value is equal to or greater than the predetermined threshold, the change detection threshold is set as the detection sensitivity. It is preferable to set the value to be increased.

  In the object change detection apparatus described above, it is preferable that the threshold setting unit sets a function based on the absolute value or the variance value as the change detection threshold.

  In the object change detection device described above, it is preferable that the change detection threshold is a fixed value determined for each detection sensitivity level.

  In the object change detection device described above, the shape information acquisition unit represents the shape information of the object from the image including the object detected by the object detection unit using a predetermined feature amount. It is preferable to acquire a feature amount distribution.

  In the object change detection device described above, the shape change information acquisition unit normalizes the feature amount distribution acquired by the shape information acquisition unit, and acquires a probability distribution corresponding to the feature amount distribution. And shape information storage means for storing the probability distribution normalized by the normalization means, and using the predetermined scale, the probability distribution before the predetermined time stored by the shape information storage means and the current probability distribution It is preferable to further comprise information acquisition means for calculating a difference between the information and the shape change information of the object.

  The object change detection method of the present invention includes an object detection step for detecting an object from an image, and a shape information acquisition step for acquiring a feature amount distribution representing shape information of the object detected in the object detection step. Shape change information acquisition for acquiring shape change information of the object indicating a time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired in the shape information acquisition step A threshold setting step for setting a change detection threshold used for detecting a change in the object based on the shape change information acquired in the step, the shape change information acquisition step, and the shape change information acquisition step. Using the shape change information and the change detection threshold set in the threshold setting step, an object change that detects a change in the object is detected. Characterized in that it comprises a detection step.

  The present invention has an effect that a change in an object around a vehicle can be detected quickly and accurately.

FIG. 1 is a block diagram illustrating an example of the configuration of the object change detection device according to the present embodiment. FIG. 2 is a flowchart showing an example of basic processing of the object change detection apparatus according to the present embodiment. FIG. 3 is a flowchart illustrating an example of details of the shape change information acquisition process according to the present embodiment. FIG. 4 is a flowchart illustrating an example of details of the object change detection process according to the present embodiment. FIG. 5 is a diagram illustrating an example of a change detection graph according to the present embodiment.

  Hereinafter, embodiments of an object change detection device and an object change detection method according to the present embodiment will be described in detail with reference to the drawings. In addition, this invention is not limited by this embodiment. In addition, constituent elements in the following embodiments include those that can be easily assumed by those skilled in the art or those that are substantially the same.

  The configuration of the object change detection device according to the present embodiment will be described with reference to FIG. FIG. 1 is a block diagram illustrating an example of the configuration of the object change detection device according to the present embodiment.

  As shown in FIG. 1, the object change detection device 1 is configured by using, for example, an automobile control computer mounted on a vehicle and is connected to a camera 2 mounted on the vehicle so as to be communicable. ing. The camera 2 is a device that captures the periphery of the host vehicle and generates image data in which the periphery of the host vehicle is projected. The camera 2 is, for example, a CCD camera or a CMOS camera that can capture an image in the visible light region or the infrared region. The camera 2 is installed at an arbitrary position capable of capturing an image of the vehicle periphery such as the front, side, and rear of the vehicle.

  The object change detection device 1 includes a control unit 12 and a storage unit 14. The control unit 12 comprehensively controls the object change detection apparatus 1 and is, for example, a CPU (Central Processing Unit). The storage unit 14 is for storing data, and is, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), or a hard disk.

  The storage unit 14 includes a shape information storage unit 14a, a threshold storage unit 14b, a shape change information storage unit 14c, and an averaged shape change information storage unit 14d.

  The shape information storage unit 14a is a shape information storage unit that stores shape information of an object extracted from image data. The shape information is data indicating the feature amount of the shape of the object. The feature amount is not limited to this, but the first feature amount using the luminance of the image data itself, the second feature amount using the edge of the image data, and the third feature amount using the color of the image data. At least one of them. The first feature amount includes, for example, at least one of luminance, luminance PCA, Hu moment, LBP, Haarlike feature, and position. The second feature amount includes, for example, at least one of SIFT, PCA, SURF, GLOH, shape context, HOG, CoHOG, FIND, and edge. The third feature amount includes, for example, at least one of RGB and Lab.

  The threshold value storage unit 14b is a threshold value storage unit that stores a change detection threshold value used for detection of a change in an object set by a threshold value setting unit 12g described later. The details of the change detection of the object will be described later. Here, the object includes at least one of moving objects such as a pedestrian, a bicycle, a motorcycle, and a vehicle that exist around the host vehicle. The object may be a part of an object such as a vehicle door. In the present embodiment, the change detection threshold is a fixed value determined for each detection sensitivity level. For example, the threshold storage unit 14b has a high-sensitivity change detection threshold value that has a value that increases the detection sensitivity so that a change in the object can be accurately detected, and a detection sensitivity that prevents excessive detection of the change in the object. A low-sensitivity change detection threshold having a decreasing value is stored. As an example, in FIG. 5 described later, the high-sensitivity change detection threshold is “0.92”, and the low-sensitivity change detection threshold is “0.88”.

  The shape change information storage unit 14c is a shape change information storage unit that stores shape change information indicating a time change of the shape information within a predetermined time. The averaged shape change information storage unit 14d is an averaged shape change information storage unit that stores averaged shape change information indicating the average of the shape change information. In addition, the storage unit 14 may include a processing target area storage unit that stores a processing target area corresponding to a part of the image area of the image data including the target.

  The control unit 12 includes an object detection unit 12a, a shape information acquisition unit 12b, a shape change information acquisition unit 12c, a threshold setting unit 12g, and an object change detection unit 12h. Here, the shape change information acquisition unit 12c further includes a normalization unit 12d, a shape information storage unit 12e, and an information acquisition unit 12f. The object change detection unit 12h further includes a shape change information storage unit 12i, a shape change averaging unit 12j, and a change detection unit 12k.

  The object detection unit 12a is an object detection unit that detects an object from an image. The object detection unit 12a detects the object by performing pattern matching or the like using data indicating an outline of the shape of the object stored in the storage unit 14 in advance.

  The shape information acquisition unit 12b is a shape information acquisition unit that acquires a feature amount distribution representing the shape information of the object detected by the object detection unit 12a. Specifically, the shape information acquisition unit 12b acquires the feature amount distribution of the object detected by the object detection unit 12a using a predetermined feature amount. Here, the predetermined feature amount is a feature amount that can be expressed as a probability distribution.

  The shape change information acquisition unit 12c acquires the shape change information of the object indicating the time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired by the shape information acquisition unit 12b. It is change information acquisition means.

  Here, the shape change information acquisition process executed by the shape change information acquisition unit 12c includes the normalization unit 12d, the shape information storage unit 12e, and the information acquisition unit 12f included in the shape change information acquisition unit 12c. Including. In the present embodiment, the shape change information acquisition unit 12c normalizes the shape information represented by the feature amount distribution acquired by the shape information acquisition unit 12b, acquires a probability distribution corresponding to the feature amount distribution, and normalizes the shape information. The obtained probability distribution is accumulated, and using the accumulated probability distribution, a difference between the probability distribution before a predetermined time and the current probability distribution is acquired as the shape change information of the object. Hereinafter, the normalization unit 12d, the shape information storage unit 12e, and the information acquisition unit 12f will be described.

  The normalization unit 12d is a normalization unit that normalizes the shape information represented by the feature amount distribution acquired by the shape information acquisition unit 12b. Specifically, the normalization unit 12d normalizes the feature amount distribution acquired by the shape information acquisition unit 12b, and acquires a probability distribution corresponding to the feature amount distribution.

  The shape information accumulation unit 12e is shape information accumulation means for accumulating shape information represented by the feature amount distribution acquired by the shape information acquisition unit 12b. Here, the shape information storage unit 12e stores the shape information normalized by the normalization unit 12d. Specifically, the shape information accumulation unit 12e accumulates the probability distribution acquired by the normalization unit 12d. That is, the shape information storage unit 12e stores the normalized shape information (probability distribution) in the shape information storage unit 14a. In the present embodiment, the shape information accumulation unit 12e may accumulate the feature amount distribution before normalization acquired by the shape information acquisition unit 12b.

  The information acquisition unit 12f uses a predetermined scale based on the normalized shape information accumulated by the shape information accumulation unit 12e, that is, the probability distribution stored in the shape information storage unit 14a, for a predetermined time ago. It is a shape change information acquisition means for calculating the difference between the probability distribution and the current probability distribution and acquiring it as the shape change information of the object. Here, the predetermined scale is a scale for measuring a difference between probability distributions by a distance or a pseudorange. The distance includes, but is not limited to, Lp norm, for example, L1 norm (Manhattan distance), L2 norm (Euclidean distance), L infinity norm (uniform norm), and the like. Further, the distance may include a Mahalanobis distance. When using this Mahalanobis distance, it is preferable to create a distribution from a plurality of past vectors p (t). The distance may include a Hamming distance. When this hamming distance is used, it is preferable to discretize numerical values to some extent. The pseudorange includes, but is not limited to, a Kullback-Leibler divergence (hereinafter referred to as KL information). Here, the KL information amount is a measure for measuring the difference between the two probability distributions P and Q, and is well known in the field of information theory.

  In the present embodiment, the shape change information acquisition unit 12c stores the feature amount distribution before a predetermined time stored in the shape information storage unit 14a and the feature amount distribution stored in the shape information storage unit 14a when the feature information distribution before normalization is stored by the shape information storage unit 12e. Normalize the current feature quantity distribution, acquire the feature quantity distribution before the predetermined time and the probability distribution corresponding to the current feature quantity distribution, respectively, and use the predetermined scale to obtain the acquired probability before the predetermined time The difference between the distribution and the current probability distribution is calculated and acquired as shape change information of the object.

  The threshold setting unit 12g is a threshold setting unit that sets a change detection threshold used for detecting a change in an object based on the shape change information acquired by the shape change information acquiring unit 12c.

  In the present embodiment, the threshold setting unit 12g calculates the absolute value of the shape change information acquired by the shape change information acquiring unit 12c, and when the absolute value is smaller than the predetermined threshold, the detection sensitivity is set as the change detection threshold. When the absolute value is greater than or equal to a predetermined threshold, the change detection threshold is set to a value that increases the detection sensitivity. Here, the predetermined threshold is set to a value capable of determining whether or not the object is substantially stationary based on the magnitude of the absolute value of the shape change information. That is, when the calculated absolute value is smaller than the predetermined threshold, the threshold setting unit 12g determines that the object is almost stationary and detects the change with low sensitivity so as to reduce the detection sensitivity. Set to the threshold value. On the other hand, when the magnitude of the calculated absolute value is equal to or greater than the predetermined threshold, the threshold setting unit 12g determines that the object is in a moving state and increases the detection sensitivity so as to increase the detection sensitivity. Set to.

  Alternatively, the threshold setting unit 12g calculates a variance value of the shape change information for the past predetermined time acquired by the shape change information acquisition unit 12c, and detects the change detection threshold when the variance value is smaller than the predetermined threshold. If the sensitivity is set to a value that decreases and the variance value is greater than or equal to a predetermined threshold value, the change detection threshold value is set to a value that increases the detection sensitivity. Here, the predetermined threshold value is set to a value that can determine whether or not the object is substantially stationary based on the magnitude of the variation indicated by the variance value of the shape change information. That is, the threshold value setting unit 12g determines that the target is almost stationary when the variation indicated by the calculated dispersion value is smaller than the predetermined threshold value, so that the detection sensitivity is lowered. To the change detection threshold. On the other hand, the threshold value setting unit 12g determines that the object is in a moving state when the variation of the calculated dispersion value is equal to or greater than the predetermined threshold value, and detects a highly sensitive change so as to increase the detection sensitivity. Set to the threshold value.

  Here, the threshold value setting unit 12g sets a function (for example, a cosine function) based on an absolute value or a variance value as a change detection threshold value. The change detection threshold set by the threshold setting unit 12g is stored in the threshold storage unit 14b.

  The object change detection unit 12h uses the shape change information acquired by the shape change information acquisition unit 12c and the change detection threshold set by the threshold setting unit 12g to detect a change in the object. It is.

  Here, the object change detection process executed by the object change detection unit 12h includes each process executed by the shape change information accumulation unit 12i, the shape change averaging unit 12j, and the change detection unit 12k. Hereinafter, the shape change information accumulation unit 12i, the shape change averaging unit 12j, and the change detection unit 12k will be described.

  The shape change information accumulation unit 12i is a shape change information accumulation unit that accumulates the shape change information acquired by the shape change information acquisition unit 12c. That is, the shape change information storage unit 12i stores the acquired shape change information in the shape change information storage unit 14c.

  The shape change averaging unit 12j averages the plurality of shape change information accumulated by the shape change information accumulation unit 12i, that is, the shape change information stored in the shape change information storage unit 14c, and obtains the averaged shape change information. This is a shape change averaging means to be acquired. The shape change averaging unit 12j stores the acquired average shape change information in the averaged shape change information storage unit 14d.

  The change detection unit 12k calculates the similarity using the averaged shape change information acquired by the shape change averaging unit 12j or the averaged shape change information stored in the averaged shape change information storage unit 14d, It is a change detection means for detecting a change in an object when the similarity is smaller than a change detection threshold set by the threshold setting unit 12g. The similarity is not limited to this, but includes cosine similarity and the like.

  The change of the object means a dangerous change of the object in the traffic environment, and includes a change of the shape at the start of the movement change of the object. Examples of changes in objects include, but are not limited to, for example, changes in which pedestrians and bicycles suddenly change their movements and come out on the road, and parallel vehicles and motorcycles suddenly change lanes and interrupt their own lanes. Changes that come in front, changes that the vehicle ahead suddenly starts a U-turn, changes that the vehicle ahead suddenly starts a left or right turn to enter a store beside the road, and the door of the parked vehicle suddenly opens Including changes such as

  In addition, the control unit 12 may include a processing target region extracting unit that extracts a processing target region from an image region including the target detected by the target detection unit 12a. The processing target area is a predetermined area that is effective for the target change detection process by the target change detection unit 12h. For example, when the object is a pedestrian, the processing target area includes an area including only the entire pedestrian or only the lower body. The processing target area extraction unit may store the extracted processing target area in the storage unit 14. In this case, the shape information acquisition unit 12 b may acquire the feature amount distribution of the target object from the processing target region extracted by the processing target region extraction unit or the processing target region stored in the storage unit 14.

  Then, with reference to FIGS. 2-5, the process performed in the target object change detection apparatus 1 mentioned above is demonstrated. In the following processing, a pedestrian will be described as an example of an object existing around the vehicle, but the present invention is not limited to this. In addition, the SIFT feature amount is described as an example of the feature amount indicating the shape of the target object, but is not limited thereto. Further, although the KL information amount will be described as an example as a scale for measuring the difference between the probability distributions P and Q, it is not limited to this.

  A basic process of the object change detection device 1 according to the present embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing an example of basic processing of the object change detection apparatus 1 according to the present embodiment.

  As shown in FIG. 2, the control unit 12 acquires image data in which the periphery of the host vehicle is projected from the camera 2 (step S10).

  The target object detection unit 12a detects a pedestrian as a target object from the image data acquired in step S10 (step S20). For example, the object detection unit 12a detects the object by performing pattern matching or the like using data indicating an outline of the shape of the object such as a pedestrian stored in the storage unit 14 in advance.

  The shape information acquisition unit 12b acquires a feature amount distribution representing the shape information of the target detected in step S20 (step S30). Specifically, the shape information acquisition unit 12b calculates a SIFT feature value as a feature value distribution v (t) representing the shape information from the object detected in step S20. Here, the SIFT feature value is well known in the field of image recognition and the like as a feature value representing a shape. Also, the SIFT feature value can be expressed as a probability distribution because it can be expressed in a histogram which edge of the image data including the object is distributed in which direction.

  The shape change information acquisition unit 12c acquires shape change information indicating the time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired in step S30 (step S40).

  Here, details of the shape change information acquisition process executed by the process of the shape change information acquisition unit 12c in step S40 will be described with reference to FIG. FIG. 3 is a flowchart illustrating an example of details of the shape change information acquisition process according to the present embodiment.

As shown in FIG. 3, the normalization unit 12d of the shape change information acquisition unit 12c calculates the L1 norm of the feature amount distribution v (t) acquired in step S30 of FIG. 2 as shown in the following formula (1). The feature amount (probability distribution) p (t) is acquired by normalizing to 1 (step S41).

  The shape information storage unit 12e of the shape change information acquisition unit 12c stores the shape information acquired in step S41 in the shape information storage unit 14a (step S42). That is, the shape information accumulation unit 12e accumulates the feature quantity (probability distribution) p (t) normalized in step S41.

The information acquisition unit 12f of the shape change information acquisition unit 12c, as shown in the following formula (2), uses the feature amount (probability distribution) p (t) accumulated in step S42 and the feature amount n frames before and the current amount. A difference d (t, n) from the feature amount (that is, shape change information in the present embodiment) is calculated (step S43). The information acquisition unit 12f calculates the difference d (t, n) as the shape change information using the KL information amount as shown in the following mathematical formula (3). In this manner, the information acquisition unit 12f acquires the shape change information of the target object using the shape information accumulated in step S42.

  Returning to FIG. 2 again, the description will be continued from the processing of step S50. The threshold setting unit 12g sets the change detection threshold “Thre” used for detecting the change in the object based on the shape change information (that is, “d (t, n)”) acquired in step S40 (step S50). ).

  In the present embodiment, the threshold setting unit 12g sets the change detection threshold by one of the following two methods.

Specifically, in step S50, the threshold value setting unit 12g obtains the absolute value “S (t)” from “d (t)” of the shape change information acquired in step S40, as shown in the following mathematical formula (4). Calculate When the absolute value is smaller than the predetermined threshold “S_thre”, the threshold setting unit 12g sets the change detection threshold to a value “Th_low” that decreases the detection sensitivity. In the present embodiment, the low-sensitivity change detection threshold is, for example, “0.88” as shown in FIG. In this way, the threshold setting unit 12g calculates the absolute value of the shape change information indicating the time change of the shape information within the predetermined time acquired in step S40, and if this absolute value is smaller than the predetermined threshold Then, it is determined that the object (particularly a pedestrian) is almost stationary, and the change detection threshold is set to a value at which the detection sensitivity decreases. On the other hand, when the absolute value is equal to or greater than the predetermined threshold, the threshold setting unit 12g sets the change detection threshold to a value “Th_high” that increases the detection sensitivity when “S_thre ≦ S (t)”. In the present embodiment, the high-sensitivity change detection threshold is, for example, “0.92” as shown in FIG.

Alternatively, in step S50, the threshold value setting unit 12g, as shown in the following formula (5), from the “d (t)” of the shape change information acquired in step S40, the variance value “S (t (t)) for the past m frames. ) ”. Then, when the variance value is smaller than the predetermined threshold value “S_thre”, the threshold value setting unit 12g sets the change detection threshold value to a value “Th_low” that decreases the detection sensitivity. In the present embodiment, the low-sensitivity change detection threshold is, for example, “0.88” as shown in FIG. In this way, the threshold value setting unit 12g calculates the variance value of the shape change information indicating the temporal change of the shape information within the predetermined time acquired in step S40, and when this variance value is smaller than the predetermined threshold value Then, it is determined that the object (particularly a pedestrian) is almost stationary, and the change detection threshold is set to a value at which the detection sensitivity decreases. On the other hand, when the variance value is equal to or greater than the predetermined threshold value, the threshold value setting unit 12g sets the change detection threshold value to a value “Th_high” that increases the detection sensitivity when “S_thre ≦ S (t)”. In the present embodiment, the high-sensitivity change detection threshold is, for example, “0.92” as shown in FIG.

  The object change detection unit 12h detects a change in the object using the shape change information “d (t, n)” acquired in step S40 and the change detection threshold “Thre” set in step S50. (Step S60).

  Here, the details of the object change detection process executed by the process of the object change detection unit 12h in step S60 will be described with reference to FIGS. FIG. 4 is a flowchart illustrating an example of details of the object change detection process according to the present embodiment. FIG. 5 is a diagram illustrating an example of a change detection graph according to the present embodiment.

As shown in FIG. 4, the shape change information accumulating unit 12 i of the object change detecting unit 12 h has the shape change information d ( t, n) is stored in the shape change information storage unit 14c (step S61). Specifically, the shape change information accumulating unit 12i accumulates the shape change information d (t, n) calculated in step S40 for one frame, and a vector u (t, n, l shown in the following equation (6). ).

The shape change averaging unit 12j of the object change detection unit 12h adds the K vectors u (t, n, l) calculated in step S61 using the following formula (7), and calculates the time average. To reduce noise (step S62). As described above, the shape change averaging unit 12j averages the plurality of shape change information accumulated in step S61, and acquires the vector u (t, K, n, l) as the averaged shape change information.

The change detection unit 12k of the object change detection unit 12h calculates the similarity cos θ using the following mathematical formulas (8) and (9) (step S63).

  As shown in FIG. 5, the change detection unit 12k determines whether the similarity cos θ calculated in step S63 is smaller than the change detection threshold “Thre” set by the threshold setting unit 12g in step S50 of FIG. Is determined (step S64). The vertical axis in FIG. 5 indicates cos θ, and the horizontal axis indicates a time frame. FIG. 5 shows a case where the time average of four samples (K = 4) is taken. As shown in FIG. 5, all four have shown a sudden change near the frame 180, and represent a state in which the pedestrian has suddenly changed movement near the frame 180. Here, in FIG. 5, when the absolute value or the variance value calculated by the threshold value setting unit 12g in step S50 of FIG. 2 is smaller than the predetermined threshold value, “S_thre> S (t)” is detected as the change detection threshold value. It is assumed that a value “Th_low = 0.88” for decreasing the sensitivity is set. On the other hand, in FIG. 5, when the absolute value or the variance value calculated by the threshold setting unit 12g in step S50 of FIG. 2 is equal to or greater than the predetermined threshold, “S_thre ≦ S (t)” is detected as the change detection threshold. It is assumed that a value “Th_high = 0.92” for increasing sensitivity is set.

  If the change detection unit 12k determines in step S64 that the degree of similarity cos θ is smaller than the change detection threshold “Thre” (step S64: Yes), the change detection unit 12k determines that there is a change in the object (step S65). Thereafter, the object change detection process is terminated. On the other hand, if the change detection unit 12k determines in step S64 that the similarity cos θ is equal to or greater than the change detection threshold “Thre” (step S64: No), the change detection unit 12k determines that there is no change in the object and continues to the object. The change detection process ends. Then, after the determination of whether or not there is a change in the object, the process proceeds to step S10 in FIG.

  As described above, according to this embodiment, it is possible to detect a change in an object around the vehicle more quickly and accurately than in the prior art. For example, in the pedestrian pop-out prediction device described in Patent Document 1 of the prior art, there is a problem that the pedestrian pop-out prediction is slower than when predicting from the posture of the pedestrian (that is, the shape of the object). However, according to the present embodiment, since a dangerous change of the object in the traffic environment (for example, a pedestrian jumps out) is detected based on the shape of the object, the change of the object around the vehicle is quickly detected. And it can detect correctly. As described above, according to the present embodiment, when a dangerous change occurs in an object around the vehicle while the driver is driving the vehicle, the driver can be notified of the danger quickly and accurately. The possibility of an accident can be reduced.

  Here, dangerous changes in objects around the vehicle can be classified into continuous changes and discontinuous changes. For example, when the object is a pedestrian, examples of the continuous change include an operation in which the pedestrian enters linearly at a constant speed from the sidewalk toward the roadway. On the other hand, as discontinuous change, from the state where the pedestrian is moving along the sidewalk, suddenly changing the direction of travel and entering the roadside, the state where the pedestrian is moving at a low speed along the sidewalk For example, an operation of suddenly moving at a high speed and an operation of suddenly moving from a state where the pedestrian is stopped. In the prior art, this continuous change (for example, an operation in which the object moves continuously, an operation in which the object is stopped from moving, and an operation in which the object decelerates) is detected by linear prediction. Although possible, discontinuous changes could not be detected quickly and accurately.

  On the other hand, the object change detection device 1 according to the present embodiment detects an object from an image, acquires a feature amount distribution representing shape information of the detected object using a predetermined feature amount, and acquires the acquired feature amount. The distribution is normalized, a probability distribution corresponding to the feature amount distribution is acquired, and the acquired probability distribution is accumulated. Then, the object change detection device 1 according to the present embodiment calculates a difference between the accumulated probability distribution before the predetermined time and the current probability distribution using a predetermined scale, and calculates the time of the shape information within the predetermined time. Acquired as shape change information indicating a change. And the target object change detection apparatus 1 of this embodiment sets the threshold value for a change detection used for the change detection of a target object based on the acquired shape change information, the acquired shape change information, and the set change detection The change of the object is detected using the threshold for operation. Specifically, the object change detection device 1 of the present embodiment accumulates the acquired shape change information, averages the accumulated plurality of shape change information, acquires averaged shape change information, and acquires the acquired average. The similarity is calculated using the modified shape change information, and when the similarity is smaller than the set change detection threshold, a change in the object is detected.

  As a result, according to the object change detection device 1 of the present embodiment, discontinuous changes that cannot be adequately handled by the prior art (for example, an operation that starts moving from a state in which the object is stopped, an operation that accelerates the object) , And the operation of changing the direction of the object) can be detected with high sensitivity. For example, according to the embodiment, as an example of the discontinuous change, a pedestrian or a bicycle suddenly changes its movement and comes out on the road. A parallel vehicle vehicle or a motorcycle suddenly changes its lane and interrupts its own lane. The vehicle in front of the vehicle suddenly starts a U-turn, the vehicle in front of the vehicle suddenly starts a left or right turn to enter a store on the road, and the door of the parked vehicle suddenly opens. When a dangerous change occurs, the driver can be notified of the danger with high sensitivity.

  Here, if changes in the object are always detected with high sensitivity, if the object (especially a pedestrian) is almost stationary, there will be a slight movement of the pedestrian and a slight shift in the tracking result of the pedestrian. On the other hand, it may react excessively and may increase false detection of pedestrian jump-out determination. The reason for this is that, when adopting the object change detection method with high detection sensitivity that takes the time average shown in FIG. 4 and reduces noise, a slight movement of a substantially stationary object is captured as a large change. This is because it may end up.

  The reason why false detection of pedestrian jump-out determination increases will be specifically described. In the object change detection method that takes a time average, an angle formed by a constant vector in an n-dimensional space and a vector that averages apparent changes (hereinafter referred to as a motion vector) is seen at the final stage. Then, when the object is stationary, a certain dimension of the motion vector suddenly has a value from around zero. As a result, the angle suddenly moves from around 0 degrees to around 90 degrees. This is because with such a mechanism, even if the object is almost stationary, a slight movement may be perceived as a large change.

  Therefore, the object change detection device 1 according to the present embodiment sets a change detection threshold value used for detecting a change in the object based on the magnitude of the shape change information indicating the time change of the shape information within the acquired predetermined time. ing. Specifically, the object change detection device 1 is substantially stationary when the absolute value based on the acquired shape change information or the variation indicated by the variance value is smaller than a predetermined threshold. It is determined that the state is in a state, and a low-sensitivity change detection threshold is set so as to lower the detection sensitivity. On the other hand, the object change detection device 1 determines that the object is in a moving state when the magnitude of the calculated absolute value or the variation indicated by the variance value is greater than or equal to a predetermined threshold value, and detects the detection sensitivity. Is set to a highly sensitive change detection threshold value. That is, the object change detection device 1 of the present embodiment determines that the object is stationary when the apparent change of the object such as a pedestrian is small, and finds an aperiodic change of the object. The change detection threshold value, which is a parameter, is changed in a direction that lowers the sensitivity for a stationary object. By doing so, it is possible to reduce the erroneous detection of the pedestrian jump-out determination and further improve the recognition accuracy of the change of the object.

1 Object Change Detection Device 12 Control Unit
12a Object detection unit
12b Shape information acquisition unit
12c Shape change information acquisition unit
12d normalization part
12e Shape information storage unit
12f Information acquisition unit
12g threshold setting part
12h Object change detection unit
12i Shape change information storage unit
12j Shape change averaging unit
12k change detection unit 14 storage unit
14a Shape information storage unit
14b Threshold storage unit
14c Shape change information storage unit
14d Averaged shape change information storage unit 2 Camera

Claims (8)

Object detection means for detecting the object from the image;
Shape information acquisition means for acquiring a feature amount distribution representing shape information of the object detected by the object detection means;
Shape change information acquisition means for acquiring shape change information of the object indicating a time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired by the shape information acquisition means. When,
Based on the shape change information acquired by the shape change information acquisition means, threshold setting means for setting a change detection threshold value used for detecting change of the object;
Using the shape change information acquired by the shape change information acquisition means and the change detection threshold set by the threshold setting means, an object change detection means for detecting a change in the object;
An object change detection device comprising: The object change detecting means includes
Shape change information storage means for storing the shape change information acquired by the shape change information acquisition means;
A shape change averaging means for averaging the plurality of shape change information accumulated by the shape change information accumulating means to obtain averaged shape change information;
When the similarity is calculated using the averaged shape change information acquired by the shape change averaging means, and the similarity is smaller than the change detection threshold set by the threshold setting means, the change of the object is determined. Change detecting means for detecting;
The object change detection device according to claim 1, further comprising: The threshold setting means includes
An absolute value of the shape change information acquired by the shape change information acquisition means or a variance value for a predetermined past time is calculated, and if the absolute value or the variance value is smaller than a predetermined threshold, the change detection threshold is detected. 3. The change detection threshold is set to a value that increases detection sensitivity when the absolute value or the variance value is equal to or greater than the predetermined threshold. The object change detection device described. The threshold setting means includes
The object change detection device according to claim 3, wherein a function based on the absolute value or the variance value is set as the change detection threshold.   5. The object change detection device according to claim 1, wherein the change detection threshold is a fixed value determined for each detection sensitivity level. 6. The shape information acquisition means includes
6. The feature quantity distribution representing shape information of the target object is acquired from the image including the target object detected by the target object detection means using a predetermined feature quantity. The target object change detection apparatus as described in any one of Claims. The shape change information acquisition means includes
Normalizing means for normalizing the feature quantity distribution acquired by the shape information acquiring means and acquiring a probability distribution corresponding to the feature quantity distribution;
Shape information storage means for storing the probability distribution normalized by the normalization means;
Information acquisition means for calculating a difference between the probability distribution before a predetermined time stored by the shape information storage means and the current probability distribution using a predetermined scale and acquiring the difference as the shape change information of the object When,
The object change detection device according to any one of claims 1 to 6, further comprising: An object detection step for detecting the object from the image;
A shape information acquisition step of acquiring a feature amount distribution representing the shape information of the object detected in the object detection step;
Shape change information acquisition step for acquiring shape change information of the object indicating a time change of the shape information within a predetermined time based on the shape information represented by the feature amount distribution acquired in the shape information acquisition step. When,
Based on the shape change information acquired in the shape change information acquisition step, a threshold setting step for setting a change detection threshold used for detecting change in the object;
An object change detection step for detecting a change in the object using the shape change information acquired in the shape change information acquisition step and the change detection threshold set in the threshold setting step;
A method for detecting an object change, comprising:

Images