JP3562278B2 - Environment recognition device - Google Patents

Description

ここで、ｉ、ｉ’はデータ集団の位置を示し、それぞれ独立に選ばれることを示している。すなわちｉ’を１からｐまでの各データ集団に対して式（１）を適用して演算を行なう。
【００４０】
各Ｓ_{ｉｉ ’}の演算値とデータ集団内の画素信号数ｎを式（２）に代入して分散、共分散Ｖ_{ｉｉ ’}を演算する。ここでｉとｉ’は同じ数字ならその画素列の分散値で、異なる数字ならそれぞれの画素列の共分散の演算値となる。
【数２】

【００４１】
演算された各分散および共分散の演算値を式（３）の固有値方程式に代入し、固有値方程式を解くと、ｐ個の固有値λが得られる。
【数３】

ｐ個の固有値λを、
λ_１≧λ_２≧…≧λ_ｋ≧…≧λ_ｐとすれば、
｛λ_ｋ｝をそれを求めたＶ_{ｉｉ ’}の固有値に対応する。ここでｋ＝１〜ｐである。
【００４２】
このとき式（４）から、
【数４】

ｌ_ｋｉ（ｌ_１１、ｌ_１２、…、ｌ_１ｐ、ｌ_２１、ｌ_２２、…、ｌ_２ｐ、ｌ_ｐ１、ｌ_ｐ２、…、ｌ_ｐｐ）を求めると、ｎ組の固有ベクトルが得られる。これによって、ｎ個の正規化主成分特徴量が得られるが、寄与率が７０〜８０％で画像の特徴認識ができるので、例えば式（５）に示すような第１正規化主成分特徴量と第２正規化主成分特徴量が用いられる。
【数５】

ここで＊は任意行の画素信号を表示する。
【００４３】
ステップ２０５では、第１正規化主成分特徴量と第２正規化主成分特徴量を軸とした新たな情報空間を定義する。第１正規化主成分特徴量と第２正規化主成分特徴量の線形表示式は変換関数となる。
ここで前画像も同じ２つの正規化主成分特徴量であるならば、情報空間は同じで、変換関数は同じ構造としパラメータが変更されることになる。
【００４４】
ステップ２０６では、撮像空間上の情報信号Ｘ_ｉｊを行ごとに式（５）に代入して第１正規化主成分特徴量Ｚ１、第２正規化主成分特徴量Ｚ２を演算し、撮像空間での撮像情報を情報空間へ変換する。図５は変換前後のデータの分布状態を示す。撮像空間Ｓでは物体ａ、ｂの明度情報ａｓ、ｂｓとその平面位置情報（Ｘ、Ｙ）が関数Ｆによって第１正規化主成分特徴量と第２正規化主成分特徴量で構成された情報空間Ｍに変換すると、その特徴に応じて情報空間Ｍでは異なる分布となる。
【００４５】
第１正規化主成分特徴量および第２正規化主成分特徴量の物理的意味付けによって認識できる。物理的意味付けができない場合、車両が走行にしたがったクラスタの移動や分解などで先行車の有無や割り込み車両の発生、走行路の混雑度を認識することができる。物理的意味付けは分析者の機知によって行なわれるが、クラスタの移動や分解が客観的に検知できる。
【００４６】
ステップ２０７では、情報空間上のデータに対し、データクラスタリング解析を行なう。データクラスタリング解析は従来と同じように行なう。例えばデータ間の距離によって幾つかのデータグルプに分類することができる。
ステップ２０８においては、データクラスタの時間的変化を分析して環境認識を行なう。
【００４７】
次に図６、図７のフローチャートにしたがってクラスタ判断部９４におけるクラスタ分析および制御信号作成部９５での制御信号の作成を説明する
まずステップ３０１において、クラスタがあるか否かを判定する。ない場合、ステップ３０２でＣＣＤカメラを含めた撮像部が故障と判定する。このとき制御信号作成部９５は先行車検出不能の信号をアダプティブクルーズコントロール装置１０に出力し車間距離管理不能の警報を出させてドライバーに注意を促す。
【００４８】
そしてクラスタがあると判定された場合は、ステップ３０３へ進み、ここではクラスタの数について判定を行なう。クラスタが２つ以上ある場合ステップ３０４へ進む。ステップ３０４では情報空間を構成する正規化主成分特徴量の原点近傍にクラスタが存在するかどうかを判定する。
原点の近傍にクラスタが存在するならば、画角が異常に大きい物体が検出されるので、ステップ３０５において超接近物体を検出したものと判断する。このとき制御信号作成部９５は注意を促す警報や減速制御を行なうよう制御信号をアダプティブクルーズコントロール装置１０に出力する。
【００４９】
そしてステップ３０４で原点近傍にクラスタがないと判定されればステップ３０６において新しいクラスタが出現したかどうかの判定を行なう。新しいクラスタが出現していない、かつステップ３０７でクラスタが消失したと判定されれば、前方車両がＣＣＤカメラ１の視界からなくなった、あるいは走行路の混雑度が軽減したと認識する。このとき制御信号作成部９５は車両が減速中なら車速を目標速度まで戻すよう制御信号をアダプティブクルーズコントロール装置１０に出力する。
ステップ３０７で、クラスタが消失していないと判定されれば走行状況の認識が不能としてステップ３０９において次の画像が入力されるまで判断を保留する。
【００５０】
そしてステップ３０３で１つのクラスタが存在すると判定されれば、ステップ３１０でクラスタが移動しているかどうかを判定する。クラスタが移動いているならば、ステップ３１３で前方に追従車がなく、景色のみの画像と判断する。
ステップ３１０でクラスタが移動してしないと判定されればステップ３１２で相対速度が極めて小さい近接物体を一定距離で追従していると判断する。
制御信号作成部９５は画像が景色のみの場合車両を目標速度で走行できるよう制御信号を出力する。先行車を追従しているならば制御特性を維持する制御信号を出力する。
【００５１】
ステップ３０６で新しいクラスタが出現したと判定されれば、ステップ３１４へ進む。ステップ３１４でクラスタが多数出現したと判定されればステップ３１５において前方の走行路の混雑度が上昇し多数の車両が検出されたと判定する。このとき制御信号作成部９５は車両が減速中ならそれを維持し、目標速度で走行するなら減速信号をアダプティブクルーズコントロール装置１０に出力する。
【００５２】
そしてステップ３１４での判定で多数のクラスタが出現していないならば、ステップ３１６においてクラスタが遠方に向かって移動しているか否かを判定する。移動する場合ステップ３１８で自車が早く前方車群に接近中と判断する。クラスタが移動していないならば先行車と自車の間に車両割り込みが発生したと判断する。
制御信号作成部９５は何れの場合も車両を減速するよう制御信号をアダプティブクルーズコントロール装置１０に出力する。
【００５３】
次に、主成分分析による画像認識の結果とレーザレーダの検出結果を照合してＣＣＤカメラ１と環境認識装置９からなる画像認識システムの作動状態を診断することを説明する。
この診断は図１に示すようにフェール監視装置１１を環境認識装置に接続して使用される。
フェール監視装置１１が画像認識システムが故障したと判断した場合、フェール信号を出しアダプティブクルーズコントロール機能を停止させ、車両を安全な操作に切り替える。
【００５４】
フェール監視装置１１はレーザレーダ２の検出値と主成分分析による画像認識の結果を入力して診断を行なう。
診断の原理は、レーザレーダ２で障害物が検出されたにも拘わらず、情報空間内でクラスタが長期間存在しないとき、検出結果の相反でＣＣＤカメラ１からの信号に異常があるとし、画像認識システムが故障と判断できる。これは時間的にある対象をある範囲の中に捉らえているならば、必ず情報空間内にデータのクラスタが存在するからである。
【００５５】
また、データクラスタが時間的、もしくは同一時間内でも情報空間内で分散して、ばらつく場合もＣＣＤカメラのトラブルか、霧などによる影響が考えられる。いずれにしても、正確に走行環境を認識できないため、フェール信号を発生する。
【００５６】
次に、フェール監視装置１１における診断の流れを説明する。
まずステップ４０１において、環境認識装置に入力されたレーザレーダの検出値からレーザレーダが障害物を検出したかを判定する。障害物が検出された場合、ステップ４０２においてクラスタ分析の結果を入力しクラスタが存在するならば、環境認識装置は良好として診断を終了する。
【００５７】
レーザレーダが障害物を検出しているにも拘わらずステップ４０２でクラスタが存在しないという検出結果となればステップ４０４でＣＣＤカメラを含めた撮像部が故障すると判断する。
そしてステップ４０１でレーザレーダが障害物を検出していない場合ステップ４０３へ進む。ここでクラスタの時間的変化があるかどうかを判定する。変化がなかった場合ステップ４０４で撮像部が故障と判断する。クラスタの時間的変化がある場合、ステップ４０５でクラスタ分布がランダムかどうかの判定を行なって診断を終了する。
ステップ４０５で判定でクラスタ分布がランダムであるとなれば４０６において撮像部故障もしくは視界不良と判断する。
【００５８】
本実施例は以上のように構成され、主成分分析を用いて走行路画像を認識するから、走行路状況の変化を誤検出することなく検知できる。例えば前方車両追従時に側方から車両が自車車線内に進入してきた場合、その変化を情報空間におけるデータクラスタの変化として素早く捉らえられる。これによってアダプティブクルーズコントロール装置は迅速に制御特性を変えることができ、走行路状況に適合した走行制御が行なえる。
【００５９】
また刻々の情報空間を求める演算が不要で、演算負担が軽減され、線形結合のみの処理で空間の変換ができる。リアルタイムで走行路状況を認識することができるとともに走行環境を選ばない効果が得られる。
本実施例では、アダプティブクルーズコントロール装置を制御対象としたが、これに限らず、アダプティブクルーズコントロール装置が装備されていない車両ではエンジンブレーキを制御の対象としてもよい。このとき、アダプティブクルーズコントロール装置による減速制御をエンジンブレーキで行なう。また車両が既にエンジンブレーキ中の場合、エンジンブレーキをより利くように制御する。
【００６０】
また、画像認識システムの作動状態をクラスタの分布と撮像空間における関係により、診断を行なうので、画像認識のデータをそのまま利用し、演算負担にならずに装置の作動状況を判別できる。
ＣＣＤカメラ１は撮像手段を構成している。
走行路状況検出装置２０、環境変化認識部９１、トリガ信号発生部９２は環境変化検出手段を構成している。
主成分分析部９３、クラスタ判断部９４は環境認識手段を構成している。
ステップ２０４、ステップ２０５は変換関数変更手段を構成している。
【００６１】
【発明の効果】
請求項１記載の発明では、位置情報を内部変数情報に変換して環境認識するとともに、環境変化検出手段が環境変化を検出し、その出力を変換関数を求めるタイミングとするから、環境に変化がないあるいはないと認められるとき、従来の変換関数を流用し、環境変化を認識するときの精度が維持されるとともに変換関数を求める演算が少なくなり環境認識の演算負担が軽減される。
また環境に変化がないとき変換関数を求めないから位置情報の変換は同一条件で行なわれ、物理的意味付けによる認識が行なえるとともに、内部変数情報上で位置情報の位置変化により高速移動体、低速移動体、固定体などの認識が可能となる。
【００６２】
請求項２〜９記載の発明では、撮像手段は車両の走行方向を撮像し、環境変化検出手段は画像に変化をもたらす状況を検出し、画像に変化が少ない場合正規化主成分特徴量を新たに求めず従来の情報空間を利用する。画像に変化があることが検出された場合のみ正規化主成分特徴量を求めるから、情報を総合的に認識でき、画像情報を正確に抽出でき、車載装置に求められる高速性や高信頼性が得られる。
【００６３】
請求項１０〜１６記載の発明では、情報空間に変換された画像情報の分布により分類されたデータクラスタの分析によって行なうから、走行環境の分析に有効で、走行路の混雑度や割り込み車両の発生、先行車への接近状態といった状況を高精度に検出できる。
【００６４】
請求項１７記載の発明では、前記環境変化検出手段にレーザレーダを用い、レーザレーダの検出結果と画像による環境認識の結果を照合して前記環境認識装置の作動状況を診断するから、装置の故障による誤認識が防止され車載装置として求められる信頼性が向上する。
請求項１８記載の発明では、作動状況の診断は、レーザレーダのデータクラスタの照合によって行なうから、走行路状況を認識する際のクラスタ分析の結果を利用でき診断のための処理が不要で、演算負担にならない効果が得られる。
【図面の簡単な説明】
【図１】実施例の構成を示すブロック図である。
【図２】環境認識装置の詳細を示すブロック図である。
【図３】環境認識するためのフローチャートである
【図４】画像空間の構成を示す図である。
【図５】画像情報が撮像空間と情報空間での表示の差異を示す図である。
【図６】クラスタ分析の流れを示すフローチャートである。
【図７】クラスタ分析の流れを示すフローチャートである。
【図８】装置診断のためのフローチャートである。
【図９】従来例を示すブロック図である。
【図１０】差分演算値と輪郭の関係を示す図である。
【符号の説明】
１ ＣＣＤカメラ
２ レーザレーダ
３ ナビゲーション装置
４ インフラ受信機
５ 照度センサ
６ アクセル開度センサ
７ 車速センサ
８ ハンドル操舵角センサ
９ 環境認識装置
１０ 変速装置コントローラ
１１ フェール監視装置
３０ データ処理装置
５２ 画像入力装置
５３ 領域分割装置
５４ 重心計算装置
５５ 移動量推定装置
９１ 環境変化認識部
９２ トリガ信号発生部
９３ 主成分分析部
９４ クラスタ判断部
９５ 制御信号作成部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an environment recognition device that recognizes a running environment of a vehicle by multivariate analysis.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, in driving a vehicle, it has been desired to automatically detect an obstacle that hinders traveling regardless of a person. In particular, if the detection of moving objects such as traveling vehicles is automatically performed, functions such as warning of excessive approach, automatic braking, and tracking can be realized, improving driving safety and reducing driving burden. can do.
[0003]
As a moving object detecting device, there is one described in, for example, JP-A-3-260813. As shown in FIG. 9, the configuration is such that a camera 51 that captures an image of a detection direction of a moving object, an image input device 52 that inputs a captured image of the camera 51 at predetermined time intervals, and an area that determines a detection target area based on the brightness distribution of the image. A dividing device 53, a center of gravity calculating device 54 for calculating the center of gravity and area of the determined detection target area, and a moving object is detected from a temporal change in the center of gravity and the area, and the moving amount of the moving object is estimated based on the change. And a moving amount estimating device 55.
[0004]
In Japanese Patent Application Laid-Open No. 64-26913, an edge is detected by performing difference processing of brightness information in image pickup information of a camera, an edge is recognized as a contour of an object, and a movement amount is calculated based on the edge information. . Furthermore, it detects changes in the center of gravity and area of a closed section surrounded by regions with different temporal contours and brightness, and compares them with changes in the center of gravity and changes in area corresponding to changes in distance over time. The tracking of an object is performed by confirming that the object is the same. Thereby, the moving speed can be obtained from the tracking of the moving object and the change of the moving object.
[0005]
However, the moving object detection disclosed in both of the above publications focuses on extracting a difference in brightness information and a closed section area as a premise. In the difference processing, as shown in FIG. 10, while calculating the difference in the brightness value while scanning the image in the horizontal direction, the position where the difference value is obtained is specified. Since the difference value is detected at a place where the difference in brightness is large, a contour is detected by scanning one screen and connecting the detection positions of each difference value with a line under a condition that the difference value is within a certain distance.
[0006]
The moving object has outline information with a large difference in brightness and internal information with a small difference in brightness. If the outline can be clearly extracted, the processing speed is improved by reducing the amount of information without handling the internal information. However, the lighting conditions surrounding the vehicle change every moment, and it is rare that the contour can be completely and clearly extracted due to the influence of disturbance light or the like. Therefore, image recognition using only contours is not suitable for detecting moving objects because of its small amount of information and poor response to environmental changes. It is hard.
[0007]
In addition, extracting a closed section directly from brightness information is susceptible to shadows and disturbance light, and the brightness changed by the disturbance light directly affects changes in the center of gravity and area. There was a problem of low recognition logic.
[0008]
In addition, there are detection methods in which internal information is added to contour information, such as template matching and texture matching. However, there is a problem that processing for preparing processing time and information on a template in advance is required.
[0009]
Against this background, in recent years, as a derivative of the technology for compressing image information, the imaging information of a camera is regarded as a distribution of brightness data, and a principal component analysis, which is one of multivariate analyses, is performed. A method has been proposed in which an image is recognized by mapping onto an information space spanned by feature amounts and analyzing the behavior of data in the information space.
[0010]
In the information space composed of the normalized principal component features, for example, the data distribution of the surrounding scenery after mapping and the data distribution of the following vehicle where the information does not change significantly on the moving object or the imaging surface interrupted in the background Are different. Therefore, a background, a moving object, an object that moves relatively little, and the like can be detected by a change in a data cluster in a space due to the normalized main component feature amount after conversion. In such a detection method, information of all images is processed, and a screen that cannot be dealt with by the above-described conventional processing such as a change in illumination can be processed.
[0011]
[Problems to be solved by the invention]
However, the image recognition based on the multivariate analysis as described above is mainly used for still images, and the following problems arise when it is used for the recognition of the driving environment of a car.
(1) If the background is stationary, it is easy to separate the converted data cluster corresponding to the background and the converted data cluster of a moving object having a new motion mixed into the background. In such a case, since the background changes as the vehicle travels, the information space based on the normalized principal component feature is not uniquely determined temporally, and therefore, its conversion function cannot be fixed.
[0012]
(2) In order to solve the above problem, the information space and the conversion function may be obtained for each image. However, the amount of calculation increases and it takes too much time. Disappears.
(3) Also, if the information space and the conversion function are different due to a change in the image, the physical meaning of the normalized principal component changes depending on the processing result, so that the meaning of the data cluster after the conversion and the It is difficult to determine a significant difference between data clusters.
In view of the above problems, the present invention obtains the information space and the conversion function only when it is recognized that there is a change in the image, and otherwise uses the conventional information space and the conversion function, and does not impair the detection accuracy and the amount of calculation. It is an object of the present invention to provide an environment recognition device in which the number is reduced.
[0013]
[Means for Solving the Problems]
Therefore, the invention according to claim 1 includes an information input unit for inputting spatial or planar position information for environment recognition,
Internal variable information applied to environment recognition,
A conversion function that analytically converts the position information into the internal variable information,
An environment recognition device having environment recognition means for converting the position information into the internal variable information using the conversion function and performing environment recognition.
Conversion function changing means;
Providing environmental change detection means,
The environment change detecting means detects that there is a change in the environment, and the conversion function changing means changes a conversion time period, a parameter, or a structure of the conversion function.
[0014]
Claims 2-9According to the invention described above, the environment recognition device is used for recognizing a traveling environment of a vehicle, the position information input unit is an imaging unit that captures an image of a traveling direction of the vehicle, and the conversion function is a normalized principal component feature amount. A conversion function, the internal variable information is an information space configured with the normalized principal component feature amount as an axis,
The environment change detecting means includes a laser radar, an illuminance sensor, and a navigation device.OrIn a configuration including any one or a combination of an infrastructure receiving device, a traveling speed sensor, an accelerator opening sensor, and a steering angle sensor,
No obstacle is detected by the laser radar,
Laser radar tracks the movement of obstacles in time,
Illuminance does not change rapidly,
No change in the road ahead is detected from the information of the navigation device and the infrastructure receiving device,
The average of the running speed does not change mathematically significantly,
The variance of the accelerator operation does not change mathematically significantly,
Steering wheel does not steer greatly
In either case, it is assumed that there is no change in the environmentThe environment recognizing means uses a conventional conversion function to convert image information into an information space spanned by normalized principal component features.I was foolish.
If the environment changes,The conversion function changing means performs principal component analysis on the image information, finds a new normalized principal component feature, updates the conversion function and the information space, and the environment recognition means uses the updated conversion function and information space.
[0015]
Claims 10 to 16In the described invention, the environment recognition is performed by analyzing a data cluster classified according to a distribution of image information converted into an information space,
When a cluster is detected in the vicinity of the origin of the normalized principal component axis constituting the information space with a plurality of clusters, it is recognized that the vehicle has a super approaching object,
If it detects that there is only one cluster and it is moving, it recognizes that there is no preceding vehicle and only the scenery is imaged,
When it is detected that the number of clusters is one and the vehicle is not moving, it recognizes that the preceding vehicle having a lower relative speed to the own vehicle is running,
If the new cluster does not appear and the conventional cluster can no longer be detected, it recognizes that the preceding vehicle has disappeared from view or that the congestion degree of the traveling road has been reduced,
When a large number of clusters are detected, the degree of congestion on the road increases, and it is recognized that there are many preceding vehicles,
When a new cluster appears and the conventional cluster detects that it is moving away, it recognizes that the vehicle is approaching the group of vehicles in front,Or
When a new cluster appears and it is detected that the conventional cluster has not moved, it is recognized that an interrupted vehicle has appeared between the preceding vehicle and the host vehicle.
[0016]
Claim 17In the described invention, a laser radar is used for the environment change detecting means, and the operation result of the environment recognition device is diagnosed by collating a detection result of the laser radar with a result of environment recognition by an image.
Claim 18According to the described invention, the diagnosis of the operating condition is performed when the laser radar detects an obstacle, when no data cluster exists on the information space, or when a temporal continuity change occurs in a planar distribution of the data cluster. When no data is found or when the distribution of the data clusters is no longer spatially correlated, the environment recognition device determines that a failure has occurred.
[0017]
[Action]
According to the first aspect of the present invention, the environment change detecting means detects the environmental change, and the conversion function changing means updates the conversion function, so that a moving environment can be recognized. Compared with recognizing a moving environment by obtaining a conversion function every time, the amount of calculation for obtaining the conversion function is greatly reduced, and image recognition can be performed using an appropriate conversion function even in an environment that changes drastically.
If the conversion function is not updated, the conversion of the position information is performed under the same conditions, and recognition based on physical meaning can be performed. In addition, recognition of a high-speed moving body, a low-speed moving body, a fixed body, and the like can be performed by temporally moving the position information. Becomes possible.
[0018]
Claims 2-9In the described invention, the imaging means as the information input means captures an image of the traveling direction of the vehicle, and the environment change detecting means includes a laser radar, an illuminance sensor, a navigation device, an infrastructure receiving device, a traveling speed sensor, an accelerator opening sensor,OrWhen a situation that causes a change in the image is detected by using a steering angle sensor or the like, the vehicle is traveling on a traveling road with no preceding vehicle or at a fixed distance from the preceding vehicleIf there is no change in the environmentBy using the conventional conversion function instead of updating the conversion function, it is possible to recognize whether or not the traveling road condition is maintained, and to use the image when there is a large change in the image due to the occurrence of an interrupted vehicle or a curved road. Is recognized using the conversion function corresponding to.
At this time, the image information is subjected to principal component analysis, and the display formula of the normalized principal component feature amount is set as a conversion function. The internal variable information uses an information space centered on the normalized principal component feature amount. As a result, the brightness information of the entire screen is used, and the influence of a change in lighting or the like is eliminated, and a highly robust environment can be recognized.
[0019]
Claims 10 to 16In the described invention, the analysis is performed by analyzing the data clusters classified according to the distribution of the image information converted into the information space, so that it is effective for analyzing the traveling environment, and the congestion degree of the traveling road, the occurrence of interrupted vehicles, and the A situation such as an approaching state can be detected with high accuracy.
[0020]
Claim 17In the invention described above, a laser radar is used for the environment change detecting means, and the operation result of the environment recognition device is diagnosed by collating the detection result of the laser radar with the result of environment recognition by an image. And reliability required as an on-vehicle device is improved.
Claim 18In the described invention, the diagnosis of the operating condition is performed by comparing the detection result of the laser radar with the result of the analysis on the data cluster. Is unnecessary, and an effect of not burdening the operation can be obtained.
[0021]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment in which the present invention is applied to recognition of a driving environment of an automobile will be described.
First, the driving environment recognition by the principal component analysis will be described.
Principal component analysis is a process of comprehensively analyzing information, finding a principal component representing the feature, converting the information into a principal component axis, and reading the feature.
[0022]
Since the brightness information of the driving environment is used for the recognition of the driving environment of the automobile, the driving path is first imaged by a camera. An imaging space is created by associating the brightness information obtained by the imaging with the image position. Thereby, brightness information is added to the two-dimensional plane position, and three-dimensional image information is obtained. The image information is divided into several data groups according to the image position, and is used as an explanatory variable.
[0023]
Linear variable information is provided to extract information from the explanatory variables. Then, in order to extract information from the variable information to minimize loss, an eigenvalue equation is used, and two eigenvalues having large values are extracted from the eigenvalue equation. Since the eigenvalues represent a large amount of information that can be extracted from the variable information, the first normalized principal component feature and the second normalized principal feature from which the largest amount of information can be extracted as variable information from the two eigenvalues having large values. The quantity is made.
The first principal component feature and the second principal component feature are spatially orthogonal to each other to form one information space, and the display formula thereof is a conversion function for converting information in the imaging space into the information space.
[0024]
If the brightness information of the imaging space is mapped to the information space spanned by the normalized principal component feature amounts, a cluster as a data group is formed according to the information characteristics. For example, the data cluster distribution after mapping of the surrounding scenery differs from the data cluster distribution of a moving object interrupted in the background or an object whose information on the imaging surface does not change significantly. Therefore, it is possible to recognize the background and the moving object, and the object that moves relatively little by the temporal change of the data cluster in the information space.
[0025]
The traveling road image changes as the vehicle travels.However, if the vehicle follows the preceding vehicle or travels straight on a traveling road with no preceding vehicle, the image changes slowly, so if the information space is fixed at this time, the information It is possible to recognize whether or not the traveling road condition continues due to the temporal change of the information in the space. For example, when a preceding vehicle or the like appears in the traveling road image, if the information space is newly obtained, the image changes greatly. Therefore, even if the information space is different, the appearance or disappearance of the preceding vehicle can be detected by the appearance or decomposition of data.
[0026]
FIG. 1 is a block diagram showing the configuration of the apparatus. The CCD camera 1 as an imaging unit is mounted in front of a vehicle (not shown) so as to capture an image of a traveling environment of the vehicle. The image signal is output to the environment recognition device 9 where image recognition is performed by principal component analysis. The environment recognition device 9 is connected to an adaptive cruise control device 10 for controlling the running of the vehicle.
[0027]
The adaptive cruise control device 10 controls the running of the vehicle at a constant speed, and is set by a driver's operation. The traveling target speed is the vehicle traveling speed at the time of setting, but in a state where the constant speed traveling cannot be performed due to the recognition result by the environment recognizing device 9, the control characteristics are changed and the traveling control adapted to the traveling road condition is performed.
That is, when the preceding vehicle is detected, the traveling speed is reduced to the target speed or less, and the control characteristic changes to the control characteristic for the inter-vehicle distance management. The control of the inter-vehicle distance may use the inter-vehicle distance corresponding to the traveling target speed as the control target. In this case, the vehicle performs the tracking operation. The adaptive cruise control device 10 is provided with an alarm function, which gives an alarm when approaching a preceding vehicle even if the vehicle speed is controlled to decelerate, and urges the driver to use a brake.
[0028]
The traveling road condition detecting device 20 is further connected to the environment recognizing device 9 and obtains a timing for changing the information space based on the detected value.
The laser radar 2 is mounted on the vehicle in a forward direction like the CCD camera 1, and the measured values are output to the image recognition device 9. In order to indicate that no obstacle is detected, a distant distance such as 200 m is output.
[0029]
The navigation device 3 detects the position of the own vehicle, and outputs travel route information from the map information to the environment recognition device 9 by the data processing device 30.
The infrastructure receiver 4 receives the broadcast of the travel route information and outputs the travel route information to the environment recognition device 9.
The illuminance sensor 5 detects the illuminance of the traveling environment and outputs it to the environment recognition device 9.
The accelerator opening sensor 6, the vehicle speed sensor 7, and the steering angle sensor 8 detect the running state of the vehicle and output the detected value to the environment recognition device 9. When the environment recognition device 9 determines that there is an image change based on the detection result, the environment recognition device 9 performs a principal component operation for newly obtaining a conversion function.
[0030]
FIG. 2 shows details of the environment recognition device 9. The environment change recognition unit 91 detects the traveling state of the vehicle from the detection signal of the traveling road condition detection device 20 or the temporal change of the signal. For example, it is determined whether there is an obstacle from the detection value of the laser radar 2. In addition, it is determined whether the laser radar 2 is performing tracking detection based on whether there is a temporal change when there is an obstacle. If tracking is being detected, there is no image change, so it is determined that the traveling road condition has not changed.
[0031]
The trigger signal generation unit 92 informs the principal component analysis unit 93 of the timing for obtaining the information space based on the recognition result of the environment change recognition unit by the trigger signal. Since the trigger signal is generated when there is a change in the image and it is necessary to obtain the information space, for example, when the following situation is detected, the conventional information space can be used, and the trigger signal is not output.
[0032]
(1) When no obstacle is detected or the obstacle is tracked by the detection value of the laser radar 2,
(2) If there is no rapid change in the curvature of the road ahead or no building appears from the navigation information or infrastructure information,
[0033]
(3) When the illuminance value does not change rapidly from the illuminance sensor 5,
(4) If the variance of the accelerator operation by the driver from the accelerator opening sensor 6 does not change mathematically significantly,
(5) When the average value of the traveling speed from the vehicle speed sensor 7 has no mathematically significant change,
(6) If the steering wheel is not steered significantly,
As a result, the conventional information space is used when the image does not change or changes slowly, and the information space is reconstructed only when a large change appears in the image.
[0034]
The principal component analysis unit 93 receives a camera image signal from the CCD camera 1 and checks whether a trigger signal is output. If there is no trigger signal, the conventional information space is used assuming that there is no significant change in the image. If there is a trigger signal, a new information space is obtained by principal component analysis from the brightness information of the image and eigenvalue calculation. The image information in the imaging space is converted into the information space by a conversion function.
[0035]
The cluster judging unit 94 forms a cluster by classifying the information distribution in the information space by a clustering process. Then, the traveling environment is recognized from the appearance or decomposition of the cluster or the change in the position.
The control signal generator outputs a control signal for changing the control characteristics of the adaptive cruise control device 10 according to the result of the image recognition. This enhances the adaptive cruise control function.
[0036]
Next, the flow of the principal component analysis in the principal component analysis unit will be described with reference to the flowchart of FIG.
First, in step 201, a captured image is input from the CCD camera 1. The captured image is composed of, for example, p pixels in the horizontal direction and n pixels in the vertical direction, and a number is given to each pixel so that the vertical and horizontal positions can be specified. The pixel signals are stored in a memory.
[0037]
In step 202, a three-dimensional imaging space S is created by associating pixel signals with pixel positions.
FIG. 4 is a plan view showing the configuration of the imaging space. Here, the pixel signal is X_ijHere, i indicates the vertical position of the image, and j indicates the horizontal position of the image. X if the pixel color is white_ij= 0, X for black_ij= 1. If it is gray, a value between 1 and 0 is used.
[0038]
In step 203, it is checked whether or not a trigger signal has been received from the trigger signal generator 92. If there is a trigger signal, it is determined that a new information space and a conversion function need to be obtained, and the process proceeds to step 204. If there is no trigger signal, the conventional information space and the conversion function can be used as they are, and the process proceeds to step 206.
[0039]
In step 204, the pixel signals in the imaging space shown in FIG. 4 are first divided in the horizontal direction for each pixel to generate p data groups (pixel columns). The data group includes n pixel signal data.
S is calculated based on equation (1) to calculate the variance and covariance of the pixel signal for each data group i._{ii '}Is calculated.
(Equation 1)

Here, i and i 'indicate the position of the data group, and indicate that they are independently selected. That is, the calculation is performed by applying the equation (1) to each data group from i 'to 1 to p.
[0040]
Each S_{ii '}And the number n of pixel signals in the data group are substituted into equation (2) to obtain the variance and covariance V_{ii '}Is calculated. Here, if i and i 'are the same numbers, they are the variance values of the pixel row. If i and i' are different numbers, they are the calculated values of the covariance of the respective pixel rows.
(Equation 2)

[0041]
By substituting the calculated values of the respective variances and covariances into the eigenvalue equation of equation (3) and solving the eigenvalue equation, p eigenvalues λ are obtained.
(Equation 3)

Let p eigenvalues λ be
λ₁≧ λ₂≧… ≧ λ_k≧… ≧ λ_pgiven that,
｛Λ_kＶ the V that sought it_{ii '}Corresponding to the eigenvalue of. Here, k = 1 to p.
[0042]
At this time, from equation (4),
(Equation 4)

l_ki(L₁₁, L₁₂, ..., l_1p, L₂₁, L₂₂, ..., l_2p, L_p1, L_p2, ..., l_pp) Yields n sets of eigenvectors. As a result, n normalized principal component features can be obtained. However, since the feature ratio of the image can be recognized at a contribution rate of 70 to 80%, the first normalized principal component feature as shown in Expression (5), for example, And the second normalized principal component feature amount.
(Equation 5)

Here, * indicates a pixel signal of an arbitrary row.
[0043]
In step 205, a new information space is defined with the first normalized principal component feature and the second normalized principal component feature as axes. The linear display expression of the first normalized principal component feature and the second normalized principal component feature is a conversion function.
Here, if the previous image also has the same two normalized principal component feature amounts, the information space is the same, the transformation function has the same structure, and the parameters are changed.
[0044]
In step 206, the information signal X in the imaging space_ijIs substituted into equation (5) for each row to calculate a first normalized principal component feature Z1 and a second normalized principal component feature Z2, and to convert imaging information in an imaging space into an information space. FIG. 5 shows the distribution state of data before and after conversion. In the imaging space S, the brightness information as and bs of the objects a and b and the plane position information (X, Y) thereof are composed of a first normalized principal component feature amount and a second normalized principal component feature amount by a function F. When converted into the space M, the distribution becomes different in the information space M according to the feature.
[0045]
It can be recognized by the physical meaning of the first normalized principal component feature and the second normalized principal component feature. If the physical meaning cannot be given, the presence or absence of a preceding vehicle, the occurrence of an interrupted vehicle, and the degree of congestion of the traveling road can be recognized by the movement or disassembly of the cluster according to the traveling of the vehicle. Although the physical meaning is determined by the knowledge of the analyst, movement or decomposition of the cluster can be objectively detected.
[0046]
In step 207, data clustering analysis is performed on the data in the information space. Data clustering analysis is performed in the same manner as in the prior art. For example, it can be classified into several data groups according to the distance between data.
In step 208, environment change is performed by analyzing a temporal change of the data cluster.
[0047]
Next, cluster analysis in the cluster determination unit 94 and generation of a control signal in the control signal generation unit 95 will be described with reference to the flowcharts of FIGS.
First, in step 301, it is determined whether or not there is a cluster. If not, in step 302, it is determined that the imaging unit including the CCD camera has failed. At this time, the control signal creation unit 95 outputs a signal indicating that the preceding vehicle cannot be detected to the adaptive cruise control device 10 and issues an alarm indicating that the following distance cannot be managed to alert the driver.
[0048]
If it is determined that there is a cluster, the process proceeds to step 303, where the number of clusters is determined. If there are two or more clusters, go to step 304. In step 304, it is determined whether or not a cluster exists near the origin of the normalized principal component feature amount forming the information space.
If a cluster exists near the origin, an object having an abnormally large angle of view is detected. Therefore, it is determined in step 305 that a super approaching object has been detected. At this time, the control signal creation unit 95 outputs a control signal to the adaptive cruise control device 10 so as to perform a warning to call attention or a deceleration control.
[0049]
If it is determined in step 304 that there is no cluster near the origin, it is determined in step 306 whether a new cluster has appeared. If it is determined that no new cluster has appeared and that the cluster has disappeared in step 307, it is recognized that the preceding vehicle has disappeared from the field of view of the CCD camera 1 or the degree of congestion on the traveling road has been reduced. At this time, if the vehicle is decelerating, the control signal generator 95 outputs a control signal to the adaptive cruise control device 10 to return the vehicle speed to the target speed.
If it is determined in step 307 that the cluster has not disappeared, it is impossible to recognize the driving situation, and the determination is suspended in step 309 until the next image is input.
[0050]
If it is determined in step 303 that one cluster exists, it is determined in step 310 whether the cluster has moved. If the cluster is moving, it is determined in step 313 that there is no following vehicle ahead and the image is only a scene.
If it is determined in step 310 that the cluster has not moved, it is determined in step 312 that an approaching object having an extremely low relative speed is following a predetermined distance.
The control signal generator 95 outputs a control signal so that the vehicle can run at the target speed when the image is only a scene. If the vehicle is following the preceding vehicle, a control signal for maintaining the control characteristics is output.
[0051]
If it is determined in step 306 that a new cluster has appeared, the process proceeds to step 314. If it is determined in step 314 that a large number of clusters have appeared, it is determined in step 315 that the degree of congestion on the front traveling road has increased and that a large number of vehicles have been detected. At this time, the control signal generating unit 95 maintains the deceleration signal if the vehicle is decelerating, and outputs a deceleration signal to the adaptive cruise control device 10 if the vehicle is traveling at the target speed.
[0052]
Then, if a large number of clusters do not appear in the determination in step 314, it is determined in step 316 whether or not the cluster has moved far away. If the vehicle is moving, it is determined in step 318 that the own vehicle is approaching the group of vehicles in front soon. If the cluster has not moved, it is determined that a vehicle interruption has occurred between the preceding vehicle and the own vehicle.
In any case, the control signal generator 95 outputs a control signal to the adaptive cruise control device 10 to decelerate the vehicle.
[0053]
Next, a description will be given of diagnosing the operation state of the image recognition system including the CCD camera 1 and the environment recognition device 9 by comparing the result of image recognition by principal component analysis with the result of laser radar detection.
This diagnosis is used by connecting the fail monitoring device 11 to the environment recognition device as shown in FIG.
If the failure monitoring device 11 determines that the image recognition system has failed, it issues a failure signal, stops the adaptive cruise control function, and switches the vehicle to safe operation.
[0054]
The failure monitoring device 11 performs a diagnosis by inputting a detection value of the laser radar 2 and a result of image recognition by principal component analysis.
The principle of the diagnosis is that, when an obstacle is detected by the laser radar 2 and no cluster exists in the information space for a long time, it is determined that the signal from the CCD camera 1 is abnormal due to the reciprocal of the detection result. The recognition system can be determined to be faulty. This is because if an object is temporally captured within a certain range, a data cluster always exists in the information space.
[0055]
Further, when the data clusters are temporally or dispersed in the information space even within the same time, the data clusters may fluctuate due to trouble of the CCD camera or fog. In any case, a failure signal is generated because the traveling environment cannot be recognized accurately.
[0056]
Next, the flow of diagnosis in the failure monitoring device 11 will be described.
First, in step 401, it is determined whether or not the laser radar has detected an obstacle from the detected value of the laser radar input to the environment recognition device. If an obstacle is detected, the result of the cluster analysis is input in step 402, and if a cluster exists, the environment recognition device determines that the condition is good and ends the diagnosis.
[0057]
If it is determined in step 402 that no cluster exists even though the laser radar has detected an obstacle, in step 404 it is determined that the imaging unit including the CCD camera has failed.
If it is determined in step 401 that the laser radar has not detected an obstacle, the process proceeds to step 403. Here, it is determined whether there is a temporal change in the cluster. If there is no change, it is determined in step 404 that the imaging unit has failed. If there is a temporal change in the cluster, it is determined in step 405 whether the cluster distribution is random, and the diagnosis is terminated.
If it is determined in step 405 that the cluster distribution is random, it is determined in 406 that the imaging unit has failed or visibility is poor.
[0058]
The present embodiment is configured as described above, and recognizes the travel road image using the principal component analysis. Therefore, it is possible to detect a change in the travel road condition without erroneous detection. For example, when the vehicle enters the lane of the host vehicle from the side when following the preceding vehicle, the change can be quickly grasped as a change of the data cluster in the information space. As a result, the adaptive cruise control device can quickly change the control characteristics, and can perform traveling control suitable for traveling road conditions.
[0059]
In addition, there is no need to perform an operation for obtaining an instantaneous information space, the operation load is reduced, and space conversion can be performed by processing using only linear combination. It is possible to recognize the traveling road condition in real time and to obtain the effect of selecting the traveling environment.
In the present embodiment, the adaptive cruise control device is controlled. However, the present invention is not limited to this. For a vehicle not equipped with the adaptive cruise control device, the engine brake may be controlled. At this time, the deceleration control by the adaptive cruise control device is performed by the engine brake. If the vehicle is already under engine braking, control is exercised to make the engine braking more effective.
[0060]
In addition, since the operation state of the image recognition system is diagnosed based on the relationship between the distribution of clusters and the imaging space, the image recognition data can be used as it is, and the operation state of the apparatus can be determined without burdening the calculation.
The CCD camera 1 constitutes an imaging unit.
The travel road condition detection device 20, the environment change recognition unit 91, and the trigger signal generation unit 92 constitute an environment change detection unit.
The principal component analysis unit 93 and the cluster determination unit 94 constitute an environment recognition unit.
Steps 204 and 205 constitute a conversion function changing means.
[0061]
【The invention's effect】
According to the first aspect of the present invention, the position information is converted into the internal variable information to recognize the environment, and the environment change detecting means detects the environment change and sets its output as the timing for obtaining the conversion function. When it is recognized that there is no or no, the conventional conversion function is diverted, the accuracy in recognizing the environmental change is maintained, and the calculation for obtaining the conversion function is reduced, so that the calculation load of the environment recognition is reduced.
In addition, when the environment does not change, the conversion function is not obtained, so that the conversion of the position information is performed under the same condition, the recognition by the physical meaning can be performed, and the high-speed moving body, Recognition of low-speed moving objects, fixed objects, and the like becomes possible.
[0062]
Claims 2-9In the described invention, the imaging means captures an image of the traveling direction of the vehicle, and the environment change detection means detects a situation that causes a change in the image. Use information space. Since the normalized principal component feature is obtained only when a change is detected in the image, the information can be comprehensively recognized, the image information can be accurately extracted, and the high speed and high reliability required for the in-vehicle device are achieved. can get.
[0063]
Claims 10 to 16In the described invention, the analysis is performed by analyzing the data clusters classified according to the distribution of the image information converted into the information space, so that it is effective for analyzing the traveling environment, and the congestion degree of the traveling road, the occurrence of interrupted vehicles, and the A situation such as an approaching state can be detected with high accuracy.
[0064]
Claim 17In the invention described above, a laser radar is used for the environment change detection means, and the operation result of the environment recognition device is diagnosed by collating the detection result of the laser radar with the result of environment recognition by an image. Is prevented, and the reliability required as an in-vehicle device is improved.
Claim 18In the invention described in the above, since the operation status is diagnosed by collating the data cluster of the laser radar, the result of the cluster analysis when recognizing the traveling road condition can be used, and the process for the diagnosis is not required, and the calculation load is not required. The effect is obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of an embodiment.
FIG. 2 is a block diagram illustrating details of an environment recognition device.
FIG. 3 is a flowchart for environment recognition.
FIG. 4 is a diagram showing a configuration of an image space.
FIG. 5 is a diagram illustrating a difference between display of image information in an imaging space and information space.
FIG. 6 is a flowchart showing a flow of cluster analysis.
FIG. 7 is a flowchart illustrating a flow of cluster analysis.
FIG. 8 is a flowchart for device diagnosis.
FIG. 9 is a block diagram showing a conventional example.
FIG. 10 is a diagram illustrating a relationship between a difference operation value and an outline.
[Explanation of symbols]
1 CCD camera
2 Laser radar
3 Navigation device
4 Infrastructure receiver
5 Ambient light sensor
6 Accelerator opening sensor
7 Vehicle speed sensor
8. Steering angle sensor
9 Environment recognition device
10 Transmission controller
11 Failure monitoring device
30 Data processing device
52 Image input device
53 area dividing device
54 Center of gravity calculator
55 Movement Estimation Device
91 Environmental Change Recognition Department
92 Trigger signal generator
93 Principal Component Analysis Department
94 Cluster judgment unit
95 Control signal generator

Claims (18)

Information input means for inputting spatial or planar position information for environment recognition,
Internal variable information for environment recognition,
A conversion function that analytically converts the position information into the internal variable information,
An environment recognition device having environment recognition means for converting the position information into the internal variable information using the conversion function and performing environment recognition.
Conversion function changing means;
Providing environmental change detection means,
An environment recognition apparatus, wherein the environment change detecting means detects a change in the environment, and the conversion function changing means changes a conversion time period, a parameter, or a structure of the conversion function. The environment recognition device is used for recognizing a traveling environment of a vehicle, the information input unit is an imaging unit that captures an image of a traveling direction of the vehicle, the conversion function is a conversion function to a normalized principal component feature amount, The internal variable information is an information space configured with the normalized principal component feature amount as an axis,
When the environment change detecting means detects that there is no change in the environment, the environment recognizing means uses a conventional conversion function to convert the image information into an information space spanned by normalized principal component features. environment recognition apparatus according to claim 1, wherein the I rows. The environment change detection means includes a laser radar,
3. The environment recognition apparatus according to claim 1, wherein when the obstacle is not detected by the laser radar or the movement of the obstacle is temporally tracked, it is determined that there is no change in the environment. . The environment change detection means includes an illuminance sensor,
The environment recognition device according to any one of claims 1 to 3, wherein when the illuminance detected by the illuminance sensor does not change rapidly, it is determined that there is no change in the environment. The environment change detecting means includes at least one of a navigation device and an infrastructure receiving device,
The method according to any one of claims 1 to 4, wherein it is determined that there is no change in the environment if the curvature of the road ahead does not change based on at least one of the information of the navigation device and the infrastructure receiving device. Environment recognition device. The environment change detection means includes a traveling speed sensor,
The environment according to any one of claims 1 to 5, wherein when the average value of the traveling speed detected by the traveling speed sensor does not mathematically change significantly, it is determined that there is no change in the environment. Recognition device. The environment change detecting means includes an accelerator opening sensor,
The method according to any one of claims 1 to 6, wherein when the variance of the accelerator operation detected by the accelerator opening sensor does not mathematically change significantly, it is determined that there is no change in the environment. Environment recognition device. The environment change detection means includes a steering angle sensor,
The environment recognition device according to any one of claims 1 to 7, wherein when the steering wheel is not largely steered, it is determined that there is no change in the environment. When the environment change detecting unit detects that there is a change in the environment, the conversion function changing unit performs principal component analysis on the image information, finds a new normalized principal component feature amount, and converts the conversion function and the information space. The environment recognition apparatus according to any one of claims 2 to 8, wherein the environment recognition unit is updated and uses the updated conversion function and information space. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
10. A vehicle according to claim 2, wherein when a plurality of clusters are detected near the origin of the normalized principal component axis constituting the information space, the vehicle is recognized as having a super approaching object. 2. The environment recognition device according to 1. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
The environment recognition apparatus according to any one of claims 2 to 10, wherein when it is detected that the number of clusters is one and the vehicle is moving, it is recognized that there is no preceding vehicle and there is only a scenery-only travel path. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
The environment according to any one of claims 2 to 11, wherein when it is detected that the number of clusters is one and the vehicle does not move, the preceding vehicle having a lower relative speed to the own vehicle is recognized as traveling. Recognition device. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
13. A vehicle according to claim 2, wherein when a new cluster does not appear and the conventional cluster cannot be detected, the preceding vehicle disappears from view or the degree of congestion on the traveling road is reduced. The environment recognition device according to 1. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
The environment recognition device according to any one of claims 2 to 13, wherein when a large number of clusters are detected, the degree of congestion of the traveling road increases and there is a large number of preceding vehicles. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
15. The method according to claim 2, wherein when a new cluster appears and the conventional cluster is detected to be moving far away, it is recognized that the own vehicle is approaching the group of vehicles in front at high speed. An environment recognition device according to any one of the preceding claims. The environment recognition is performed by analyzing a data cluster classified according to a distribution of the image information transformed into the information space,
16. The method according to claim 2, wherein when a new cluster appears and the conventional cluster is detected as not moving, it recognizes that an interrupted vehicle has appeared between the preceding vehicle and the own vehicle. Environmental recognition device as described. 2. The environment according to claim 1, wherein a laser radar is used as said environment change detecting means, and a detection result of said laser radar is compared with a result of environment recognition based on an image to diagnose an operation state of said environment recognition device. Recognition device. The diagnosis of the operating condition may be such that when the laser radar detects an obstacle, no data cluster exists in the information space, or a temporal continuity change is not found in the planar distribution of the data cluster, or data 18. The environment recognition apparatus according to claim 17, wherein when any of the distributions of the clusters has no spatial correlation, the environment recognition apparatus is determined to be faulty.

Images