JP5724955B2 - Object detection apparatus, information processing apparatus, and object detection method - Google Patents

Description

  The present invention relates to an object detection apparatus that detects an object with two or more sensors.

  Technologies for detecting and avoiding obstacles and reducing damage are known. As sensors for detecting obstacles, there are mainly radar sensors that use laser devices and camera sensors that take images. However, there are attempts to improve detection accuracy by mounting both sensors. In general, it is said that a radar sensor has high accuracy in distance and relative speed, and a camera sensor has high accuracy in orientation. The device on the control side (hereinafter simply referred to as the control device) digitizes the detection reliability of the same object by each sensor, extracts only the detection result of one sensor, weights both detection results, etc. Thus, the object is detected with high accuracy.

Also, the radar sensor and the camera sensor often have different detection ranges due to sensor performance.
FIG. 1 is an example of a diagram schematically illustrating detection ranges of a radar sensor and a camera sensor. As shown, the camera sensor has a wider detection range. A technique for utilizing the position of the object for driving support even when the object is detected only by the camera sensor by using the difference in the detection range is considered (for example, see Patent Document 1). .)

  In Patent Document 1, when an object detected by a radar sensor disappears, it is determined whether or not the object has been moved to an object detection range covered by the camera sensor based on the disappeared position and the moving direction of the object. An object detection device is disclosed in which information immediately before disappearance of an object detected by a sensor is referred to as initial information, and a camera sensor continues to detect the object.

JP 2007-272441 A

  However, since the object detection device disclosed in Patent Document 1 determines whether or not to switch only from the detection range of each sensor, there is a problem that the sensor is switched even in a scene that should not be switched in an actual traveling environment. .

  FIGS. 2A to 2C are diagrams illustrating an example of a scene of a traveling environment that should not be switched from a laser sensor to a camera sensor. 2A shows an image captured by the camera sensor, FIG. 2B shows an object detected by the radar sensor and the camera sensor (H indicates the detection result of the camera sensor, and + indicates the detection result of the radar sensor. FIG. 2C shows a fusion target obtained by fusing (detecting) the detection results of the radar sensor and the camera sensor. As shown in FIG. 2B, a total of three targets a to c are detected, but the target a is detected only in the detection range of the camera sensor.

  In such a scene, since the target a exists only in the detection range of the camera sensor, the target a is detected only by the camera sensor when switched to the camera sensor. For this reason, the control device must support driving based on target information with low reliability by only one sensor. For example, the scene shown in the figure is a scene that may occur when the host vehicle passes a target a at a close distance. In a passing scene at a close distance, the possibility of contact with the target a is low. Therefore, it is not necessary to make a collision determination by switching the detection of the target a only to the camera sensor. If the collision is determined only by the camera sensor, there is a possibility that the determination accuracy is poor (there is a possibility that it will collide with no collision). That is, in a passing scene at a close distance, the possibility of contact with the target a is low, so it can be said that it is preferable to use a radar sensor outside the detection range rather than switching to the camera sensor.

  On the other hand, a radar sensor has the property that it is difficult in principle to detect a target at a low vehicle speed and a close range. For this reason, even if a target exists in the detection range of the radar sensor and the camera sensor, there is a scene where the detection result of the camera sensor should be given priority over the radar sensor.

  2D to 2F are diagrams illustrating an example of a scene where the detection result of the camera sensor should be prioritized over the radar sensor. 2D shows an image taken by the camera sensor, FIG. 2E shows a target detected by the radar sensor and the camera sensor, and FIG. 2F shows a detection result of the radar sensor and the camera sensor. Each fusion target is shown. This scene is a scene in which the preceding vehicle is stopped at a close distance with a large lap rate (hereinafter referred to as a close-range object stationary scene). At such a close distance, the detection result of the radar sensor is low in accuracy (for example, there may be a distance jump where the distance is not continuous), so the control device should make a collision determination based on the detection result of the camera sensor. preferable.

  In order to make a collision determination based on the detection result of the camera sensor in a close-up object still scene, it is considered that the safe side can be controlled by adopting the shorter distance between the radar sensor and the camera sensor. However, by simply adopting the closer distance, a collision determination is made with information captured by only one sensor for all targets. For example, there is a possibility of performing a collision determination using only a camera sensor in a passing scene at a close distance.

  In view of the above problems, an object of the present invention is to provide an object detection apparatus capable of appropriately selecting a sensor to be used for control according to a scene in an object detection apparatus that detects a target with a plurality of sensors.

The present invention provides a first sensor for detecting first target information including a first distance to a target and a first lateral position of the target, a second distance between the target and the target. A second sensor that detects second target information including the second lateral position and has a wider target detection range than the first sensor; to target, said first lateral position and the second lateral position of the same degree is detected for a predetermined time or more, the and the first distance or less the second distance threshold, and the own vehicle is running straight If the scene specifying condition is met, the data selection means for selecting the smaller one of the first distance and the second distance as the distance to the target, and the collision with the target based on the distance selected by the data selection means then when it is determined, the drive assist means for performing a driving support, those wherein the second sensor detects targets Existence probability calculation means for calculating the existence probability so that the existence target becomes higher as the same target is continuously detected, and when the scene specifying condition is not satisfied, the driving support means calculates the existence probability. If the existence probability calculated by the means is equal to or higher than a threshold value, driving assistance is performed according to a collision determination result with the target based on the second target information .

  In the object detection apparatus that detects a target with a plurality of sensors, it is possible to provide an object detection apparatus that selects a sensor that has priority according to a scene.

It is an example of the figure which shows typically the detection range of the target of a radar sensor and a camera sensor. It is an example of the figure explaining the passing scene in the short distance, and the short distance object still scene. It is an example of the figure explaining the general | schematic characteristic of an object detection apparatus. It is an example of the schematic block diagram of an object detection apparatus. It is an example of the functional block diagram of system ECU. It is an example of the figure explaining the calculation factor of an image fusion presence probability. It is an example of the figure explaining the calculation method of a stationary object presence probability. It is an example of the flowchart figure which shows the procedure in which a sensor fusion part calculates a fusion reliability etc. FIG. It is an example of the flowchart figure which shows the operation | movement procedure of system ECU.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings.
FIG. 3 is an example of a diagram illustrating schematic features of the object detection device of the present embodiment. The object detection device calculates the existence probability of the target (image fusion existence probability, stationary object existence probability) from the detection results of the radar sensor and the camera sensor.

S10: The object detection device determines whether or not the detection results (target information) of the two sensors are fused and are stable and straight ahead. Stable / straight traveling means that the vehicle travels straight with almost no change in vehicle speed. The term “fusion” means that the detection results of the camera sensor and the radar sensor coincide with each other to the same extent.
S20: If the determination in step S10 is Yes, the object detection device sets the close distance correction flag to ON.
S30: It is determined whether the close distance correction flag is ON or OFF.
S40: When the close distance correction flag is ON, the object detection device adopts the smaller one of the radar distance and the image distance as the fusion selection distance. The radar distance is the distance detected by the radar sensor, and the image processing is the distance detected by the camera sensor.
S50: Also, the object recognition apparatus determines the image existence probability for collision determination, which has the larger image fusion existence probability and stationary object existence probability described later.

  First, it is determined that the fusion is performed in step S10, that both sensors detect the target with good accuracy (the target is surely present and the position accuracy is high). means. Stable and straight traveling means that the target has not suddenly appeared in the detection range of the camera sensor from outside the detection range of the camera sensor. Therefore, when the closest distance correction flag is ON, by adopting the smaller one of the radar distance and the image distance as the fusion selection distance, it is possible to control the safe side with respect to a target that is surely present.

  For example, when the preceding vehicle is stopped at a close distance with a large lap rate, the close distance correction flag is turned ON by the determination at S10. In this case, since the object detection device determines the collision based on the fusion selection distance, the collision determination is performed based on the closer of the radar distance and the camera distance. Therefore, even if the radar sensor is not good at detecting a target at a low vehicle speed and close range in principle, it is possible to make a collision determination on the safe side.

  Also, for example, in a passing scene at a close distance, a target may enter from outside the detection range of the camera sensor when the driver performs a steering operation. In this case, the close distance correction flag remains OFF from the condition of step S10. Therefore, the object detection apparatus performs the collision determination based on the target information detected only by the camera sensor, not the fusion selection distance, as usual. Since the stationary object existence probability of a target that has entered from outside the detection range does not increase immediately (because the threshold value is not exceeded), there is almost no collision determination based on target information detected only by the camera sensor.

  In this way, by introducing the close distance correction flag, it is possible to suppress the collision determination based on the detection result of one sensor for all targets, and the fusion selection distance only for targets that are likely to exist. Can control the in-vehicle device (the camera sensor can compensate for the area that the radar sensor is not good at). Moreover, even if the detection range of the camera sensor is wider, the collision determination by only the camera sensor can be suppressed.

[Configuration example]
FIG. 4 shows an example of a schematic configuration diagram of the object detection apparatus of the present embodiment. The object detection device 100 is controlled by a system ECU (Electronic Control Unit) 14 and mainly includes a radar sensor device 11, a camera sensor device 12, a vehicle information acquisition device 13, and an operation device 15. These devices or ECUs are communicably connected to each other via an in-vehicle LAN such as a CAN (Controller Area Network).

<Radar sensor device>
The radar sensor device 11 includes a radar sensor 21 and a radar ECU 22. The radar sensor 21 is disposed in the center of the front of the vehicle, such as the front grille of the vehicle, and emits millimeter-wave radio waves at a predetermined angle (for example, 10 degrees left and right with the front as the center). A millimeter wave reflected by a target existing in this range is received. The radar sensor 21 transmits, for example, FMCW (Frequency Modulated Continuous Wave) for obtaining a distance from a frequency difference (beat frequency) between a transmission wave and a reception wave. Alternatively, a pulse laser may be used.

  The radar sensor 21 has one transmission antenna and N reception antennas. The millimeter wave radar sensor 21 switches the reception antenna in a time division manner with a switch while transmitting a millimeter wave that rises at a constant speed and descends at a constant speed from the transmission antenna. The radar sensor 21 generates a beat signal of the transmission signal and the reception signal for each reception antenna by mixing the transmission signal and the reception signal with a mixer. The beat signal is subjected to FFT analysis, for example, and the distance and relative speed from the target can be obtained from the beat frequency when the transmission frequency is increased and the beat frequency when the transmission frequency is decreased.

  The direction of the target can be obtained by performing, for example, DBF (Digital Beam Forming) processing on the beat signal of each receiving antenna. The DBF processing is a digital circuit that realizes a phased array antenna that calculates the azimuth of a target from a phase difference by utilizing the fact that the phase of a received wave received by each antenna differs according to the azimuth of the target. In the DBF processing, the antenna directivity can be formed by arbitrarily changing the phase and amplitude by calculation. The orientation may be obtained by MUSIC analysis or Capon analysis other than DBF.

  By knowing the azimuth and distance, the lateral position of the target (the central position in the width direction of the target with reference to the lateral central position of the host vehicle) can be obtained. The radar sensor 21 transmits the distances, relative velocities, and azimuths of all the targets thus detected to the radar ECU 22.

  The radar ECU 22 includes a microcomputer including a CPU, a ROM, a RAM, an input / output I / F, a CAN controller, an A / D conversion circuit, an IC for a specific function, a power supply circuit, and the like. The other ECUs have the same configuration. In the radar ECU 22, the CPU executes a program stored in the ROM, and performs stationary object / preceding vehicle determination and relative position / speed calculation.

  In the relative position / speed calculation, the lateral position and the lap rate are calculated from the distance and direction, or the speed of the target with respect to the road surface is calculated.

  In the stationary object determination of the stationary object / preceding vehicle determination, it is determined whether or not the target is stationary with respect to the road surface. For example, if the relative speed with respect to the target can be considered to be approximately the same as the vehicle speed with respect to the road surface of the host vehicle, the target is determined to be a stationary object. In the preceding vehicle determination, for example, the speed of the target is calculated from the relative speed with respect to the target and the vehicle speed relative to the road surface of the host vehicle, and a target traveling at a speed equal to or higher than a predetermined value is set as a candidate for the preceding vehicle. Then, among the candidate targets, a target whose direction is regarded as the front direction of the host vehicle is determined as the preceding vehicle.

<Camera sensor device>
The camera sensor device 12 includes a camera sensor 23 and a camera ECU 24. The camera is a stereo camera capable of acquiring distance information, but may be a monocular camera as long as the distance information can be acquired. The camera is arranged, for example, on a rearview mirror with the optical axis facing the front of the vehicle. The camera has a right camera and a left camera that are spaced apart by a predetermined distance, and each camera is a photoelectric conversion element such as a CCD or a CMOS. The right camera and the left camera capture images almost simultaneously, and output image data to the camera ECU 24 at a predetermined cycle (for example, 30 to 60 fps).

  Manufacturing errors, lens distortions, focal length errors, and the like between the left and right cameras are calibrated at the time of vehicle manufacture and shipment. Due to the calibration, the image data of the right camera and the left camera has only a difference corresponding to the parallax.

  The camera sensor 23 calculates a parallax for the same target photographed by the left and right cameras based on the correlation between the two images photographed by the right camera and the left camera, respectively. Specifically, the image data of the left camera is fixed, and the target pixel is sequentially selected. Then, for each pixel block of a predetermined size centered on the pixel of interest, the pixel block at the corresponding position in the right camera image data is extracted, and the most correlated shift amount is determined while shifting pixel by pixel. . The correlation value may be the sum or the sum of squares of the absolute values of the luminance differences in the pixels of the two pixel blocks. For each position shifted up and down and left and right, the shift position with the smallest correlation value is obtained as the shift amount of the pixel of interest. By performing this with attention paid to all pixels, a shift amount can be obtained for each pixel.

If the parallax (the number of shifted pixels) is n, the focal length of the lens is f, the distance between the right camera and the left camera is m, and the pixel pitch is d, the distance L to each pixel is calculated from the following equation.
L = (f × m) / (n × d)
It can be presumed that a single target is shot on consecutive pixels at the same distance. If the distance is known, the azimuth of the target can be obtained from the focal length f and the target position in the image. The relative speed can be obtained by performing a differentiation process on the distance information.

  Thus, both the radar sensor device 11 and the camera sensor device 12 can obtain an orientation, a relative speed, and a distance (hereinafter, these are referred to as target information). However, since the radar sensor 21 and the camera sensor 23 have different detection ranges (generally, the camera sensor 23 is wider), only the camera sensor 23 may detect the target depending on the position of the target. Further, even when the same target is detected in both detection ranges, the distance, the relative speed, and the direction do not always match.

<Vehicle information acquisition device>
The own vehicle information acquisition device 13 includes a yaw rate / G sensor 25, a wheel speed sensor 26, and a steering angle sensor 23. The yaw rate / G sensor 25 is a sensor that can detect acceleration in two directions (front-rear direction and left-right direction) and one yaw rate. For example, the acceleration and yaw rate are detected by taking out the change in electrostatic capacitance of a plurality of minute vibration parts by an electric circuit, which is created by the MEMS technology.

  The wheel speed sensor 26 is, for example, a sensor that detects the rotation of the rotor arranged in each wheel as a wheel pulse by a vehicle-side sensor taking out from a change in magnetic flux. The rotational speed is obtained from the number of wheel pulses per unit time, and the vehicle speed can be obtained by further considering the tire diameter.

  The steering angle sensor 23 is a sensor that detects the rotation angle of the steering shaft. There are various detection principles. For example, S pole and N pole magnetic bodies are arranged on the steering shaft side, the periphery of the steering shaft is enclosed in a ring shape, and a change in magnetism is detected on the ring side. Thus, the rotation angle is detected.

<Operating device>
The operation device 15 is various in-vehicle devices used for driving support for avoiding abnormal approach to the target. The operation device 15 operates when it receives an operation command from the system ECU 14.

  Examples of the actuation device 15 include an alarm buzzer and an automatic brake device. The alarm buzzer alerts the driver that the target is approaching by blowing the buzzer on the meter panel. The automatic brake device is a device that is realized by a brake ECU and a brake actuator and brakes the vehicle without the driver depressing a brake pedal. Specifically, the brake ECU supplies the braking hydraulic pressure generated by the pump to the wheel cylinder by controlling the opening and closing of the valve of the brake ACT, and brakes the wheel independently for each wheel.

<System ECU>
The system ECU 14 performs a collision determination calculation with a target, a close distance correction calculation, and the like.
FIG. 5 shows an example of a functional block diagram of the system ECU 14. The system ECU 14 is implemented by the CPU executing a program and cooperating with hardware. The sensor fusion unit 31, the image fusion existence probability calculation unit 34, the stationary object existence probability calculation unit 35, the closest distance correction determination unit 32, and the correction process The unit 33 and the collision determination unit 36 are included.

  The sensor fusion unit 31 includes target information detected by the radar sensor device 11 (hereinafter referred to as target information L for distinction) and target information detected by the camera sensor device 12 (hereinafter referred to as target information C for distinction). Integrated). Integration refers to weighting the target information L and C to generate one target information F, or selecting one of the target information L and C. Further, the sensor fusion unit 31 determines whether or not the target information L and C are fused, and calculates the fusion reliability.

  The system ECU 14 calculates the image existence probability. The image existence probability is one of the conditions used for collision determination, and is a parameter that represents the probability that the target exists. The activation device 15 does not operate unless the image existence probability is equal to or higher than the threshold value. The image existence probability is the greater of the image fusion existence probability and the stationary object existence probability. If fusion has occurred, it is expected that the image presence probability also exceeds the threshold.

  The image fusion existence probability calculating unit 34 calculates the image fusion existence probability, and the stationary object existence probability calculating unit 35 calculates the stationary object existence probability.

  The close distance correction determination unit 32 determines for each target whether the close distance correction flag is set to ON or OFF. When the close distance correction flag is ON, the correction processing unit 33 determines the smaller one of the radar distance detected by the radar sensor 21 and the image distance detected by the camera sensor 23 as the fusion selection distance. Further, when the close-range correction flag is ON, the correction processing unit 33 determines the larger one of the image fusion existence probability and the stationary object existence probability as the image existence probability.

  The collision determination unit 36 performs the collision determination with the target based on the presence probability of the target and the target information (and TTC (Time To Collision) calculated from the target information). When the target information L and C are fused, the collision determination is performed based on the fused target information F when the image fusion existence probability is equal to or higher than a threshold value. When only the target information C is detected, the collision determination is performed based on only the target information C when the stationary object existence probability is equal to or higher than the threshold value. When only the target information L is detected, the collision determination is performed based only on the target information L (the existence probability may or may not be used).

  In the present embodiment, when the fusion is performed, the close distance correction flag may be ON or OFF, but when the close distance correction flag is OFF, the same as when the fusion is performed. The case where the close distance correction flag is ON will be described later.

<Image fusion existence probability>
First, the image fusion existence probability will be described. The image fusion existence probability is the existence probability of a target calculated based on how similar the target information L is to the target information C. That is, it is estimated that the higher the image fusion existence probability, the higher the possibility of fusion.

  FIG. 6A is an example of a diagram for explaining the calculation factor of the image fusion existence probability. In this figure, there are columns of “millimeter wave detection status”, “overlapping state”, and “image fusion existence probability increase / decrease value”.

  The “millimeter wave detection status” and the “overlapping state” are conditions for determining the “image fusion existence probability increase / decrease value”. The “millimeter wave detection status” is a detection state of the target by the radar sensor 21. In the detection of a target by the radar sensor 21, the same target cannot always be detected by each radar scan due to the influence of multipath or the like. For example, the radar sensor 21 performs FFT processing on the beat signal to set the peak frequency to the beat frequency, but the higher the peak value (power value), the higher the reliability of target detection. For this reason, for example, two threshold values are prepared, and when the higher threshold value is exceeded, the millimeter wave detection status is determined as “good”, and when the lower threshold value is exceeded, the millimeter wave detection status is determined as “normal”. . If the lower threshold is not exceeded, the millimeter wave detection status is “no detection”.

  “Overlapping state” indicates whether or not the horizontal positions of the target information C and the target information L overlap. As shown in FIG. 6B, the target of the target information C has a horizontal width, whereas the horizontal position (azimuth) of the target information L is determined by one point in principle. For this reason, the “overlapping state” is “within the image range” in which the horizontal position of the target information L is detected within the horizontal width of the target information C, and within a predetermined distance from the right or left end of the horizontal width of the target information C. There is “out of image range” in which the lateral position of the mark information L is detected. A plurality of target information L may be associated with one target information C.

  The “image fusion existence probability increase / decrease value” is determined by the relationship between the “millimeter wave detection status” and the “overlapping state”. That is, if the “millimeter detection status” is “good” and the “overlapping state” is “within image range”, it can be estimated that there is a high possibility of fusion, so the “image fusion existence probability increase / decrease value” is “+ ΔQ”. It is. ΔQ is a design value of about several percent to several tens of percent. If the “overlapping state” is “out of image range”, the possibility of fusion is slightly reduced, so the “image fusion existence probability increase / decrease value” is “+ ΔQ / 2”. The same can be considered when the “millimeter detection status” is “normal”.

  On the other hand, when the “millimeter detection status” is “no detection”, even if the peak value detected in the past is not detected in the next radar scan, it is unnatural that the target suddenly exists. Moreover, a peak may be detected again. For this reason, the “image fusion existence probability increase / decrease value” is gradually reduced. That is, if the “millimeter detection status” is “good” or “normal” and the previous “overlap state” is “within the image range” until the previous time, the “image fusion existence probability increase / decrease value” is “−ΔQ / 3”. is there. If the previous “overlap state” is “out of image range”, “−ΔQ / 2”.

  In this manner, the image fusion existence probability can be gradually increased or decreased according to the combination of the “millimeter detection status” and the “overlapping state”. The image fusion existence probability starts when, for example, the radar sensor 21 first detects the target information L with a high threshold value, and starts with a predetermined initial value (for example, 20 to 40, etc.) and detects with a low threshold value. The initial value starts from zero.

<Fusion reliability>
The fusion reliability is obtained from a period during which it is determined that the target information L and the target information C of the target are fused, for example. For example, if the target information L and the target information C are considered to be the same for 1 second, 20%, 40% for 2 seconds, 60% for 3 seconds, 80% for 4 seconds, 100 for 5 seconds or more %. In addition, the fusion reliability may be calculated by the number of times that the target information L and the target information C are considered to be the same.

  Further, the fusion reliability is decreased according to a period in which it is determined that the fusion is not performed. The increase amount and the decrease amount with respect to time may be the same or different. For example, by making the decrease amount smaller than the increase amount, it becomes easier to detect that the fusion has occurred in the past. The fusion reliability of a target that has never been fused is zero.

  In addition to the fusion reliability, when the target information L and the target information C can be regarded as the same at present, the fusion state = ON.

<Probability of stationary objects>
FIG. 7 is an example of a diagram illustrating a method for calculating the stationary object existence probability. The stationary object existence probability is a target existence probability calculated from the target information C alone. “Setting example of stationary object existence probability” defines “reset condition”, “down condition”, and “up condition”. The “reset condition” is a condition for setting an initial value for the stationary object existence probability, the “down condition” is a condition for decreasing the stationary object existence probability, and the “up condition” is the stationary object existence probability. This is a condition for decreasing. Either condition can be set as appropriate, but “reset condition” indicates that a new target has been detected, “down condition” indicates that the target is no longer detected, and “up condition” continues to detect the same target. What has been done is a condition.

  The stationary object existence probability calculating unit 35 determines whether or not the “reset condition”, the “down condition”, and the “up condition” are satisfied, and calculates the stationary object existence probability with a value determined by the “stationary object existence probability”. New setting or update. Note that the stationary object existence probability is processed so as not to exceed the range of 0 to 100%.

<Fusion reliability calculation procedure>
FIG. 8A is an example of a flowchart illustrating a procedure for the sensor fusion unit 31 to calculate the fusion reliability and the like. The system ECU 14 periodically receives the target information L and the target information C. There may be a plurality of target information L and target information C, respectively.

  The image fusion existence probability calculation unit 34 calculates the image fusion existence probability for each target information L (S110). FIG. 8B is an example of a diagram schematically showing the target information L and the target information C. The radar sensor 21 detects two targets, and transmits the target information L1 and L2 to the system ECU 14. The camera sensor 23 detects three targets and transmits the target information C1, C2, and C3 to the system ECU 14.

  Since the detection range of the camera sensor 23 is wider than the detection range of the radar sensor 21, paying attention to the target information L1 and L2, the image fusion existence probability with each of the target information C1 to C3 is calculated. Therefore, the image fusion existence probability calculation unit 34 compares the target information L1 and the target information C1 to C3, respectively, and calculates the image fusion existence probability as described in the image fusion existence probability calculation method. The same applies to the target information L2.

  Next, the sensor fusion unit 31 calculates the fusion reliability for each target information L (S120). For example, the sensor fusion unit 31 compares the target information C with the target information C closest to the target information L. If the target information L exists within the threshold from the lateral width of the target information C, the sensor information L And the target information C are the same (if they are fused). In the example of the figure, it is determined that the target information L1 and the target information C1 are fused. It may be determined whether or not fusion is performed by comparing not only the lateral position but also the azimuth and relative velocity.

  On the other hand, in the scene where the detection accuracy of any one or more of the radar sensor 21 and the camera sensor 23 decreases, the target information L and C of the same target may not be regarded as the same. In the example of the figure, the target information L2 does not exist within the width threshold of the target information C2, so it is determined that the target is not fused. Note that the image fusion existence probability is also calculated for the target information L2 based on the image fusion existence probability calculation method.

  The sensor fusion unit 31 integrates the target information L and the target information C with respect to the target information L and C determined to be fused, and generates the target information F (S130). At the time of integration, for example, the distance and relative speed are given a large weighting α to the target information L and a small weighting β to the target information C (α + β = 1 α> β). The azimuth (lateral position) is given a small weighting β to the target information L and a large weighting α to the target information C. Alternatively, the distance and relative speed may be extracted from the target information L, and the azimuth (lateral position) may be extracted from the target information C.

  Even when the target information L and C are integrated in this way, when the close distance correction flag is ON, the target information L and C are fused target information so that a sensor fusion selection distance described later can be taken out. F.

  Next, the stationary object existence probability calculating unit 35 calculates the stationary object existence probability for each target information C (S140). In FIG. 8B, since there is three target information C of target information C1 to C3, the stationary object existence probability is calculated for each.

<Close distance correction ON condition>
The close distance correction determination unit 32 determines whether or not a condition for turning on or off the close distance correction flag is satisfied. The ON condition of the close distance correction flag is a determination condition as to whether or not the safe side control is to be performed using the closer one of the radar distance and the image distance.

The close distance correction flag ON condition is, for example, the following condition. When all the conditions are satisfied, the close distance correction determination unit 32 sets the close distance correction flag to ON.
・ Fusion reliability = 100%
・ Distance ≦ 2.5m (conformity value)
・ Self-vehicle speed> 5km / h (conformity value)
・ Curve R> 1000m (conformity value)
・ Steering angle <90 ° (conformity)
・ Fusion state = ON
In addition, a suitable value means that it is a value which can be optimized suitably, and a specific numerical value is not limited to the above. Further, the determination distance may be either the radar distance or the image distance because the fusion reliability is 100%.

The close distance correction flag ON condition is intended
-For targets at close range, if the past fusion reliability is sufficient, and if the target is stable and going straight, the collision judgment on the safe side is performed.
・ If this condition is not met, the collision is determined as before.

  It should be noted that the ON condition of the close distance correction flag may not include “fusion state = ON”. This is because even if the fusion is not currently performed, if the fusion reliability is 100%, it can be estimated that the reliability of the target information C is high. As a result, the target information C can compensate for the low vehicle speed and the close distance that the radar sensor is not good at even if the fusion is not currently performed.

The close condition correction flag OFF condition refers to, for example, the following conditions. When all the conditions are satisfied, the close distance correction determination unit 32 sets the close distance correction flag to OFF.
・ Distance> 5m (conformity value)
・ Steering angle> 180 ° (conformity value)
In other words, when it is certain that the distance from the target is not close or the vehicle is not traveling straight, the close distance correction flag is turned off. This is because if the distance from the target is not close, the detection accuracy of the radar sensor 21 is sufficient, and the need for driving assistance is reduced even when the fusion reliability is 100% when the vehicle is not traveling straight.

  By determining whether the close distance correction flag is ON / OFF in this way, it is possible to control the safe side when the vehicle is fused and the vehicle is stable and straight ahead. Simply, if the fusion selection distance is always adopted, collision detection is performed with low reliability information captured by only one sensor for all targets, provided that the close distance correction flag is ON. This makes it possible to control the safe side only for targets that are likely to exist. Further, since the close distance is used as a condition, the camera sensor 23 can cover a low-speed / close-range target that the radar sensor 21 is not good at.

[Operation procedure]
FIG. 9 shows an example of a flowchart showing the operation procedure of the system ECU 14. The procedure of FIG. 9 is repeatedly executed when the power of the radar sensor 21 and the camera sensor 23 is turned on, for example.

  First, the close distance correction determination unit 32 determines whether or not the close distance correction flag ON condition is satisfied (S210). The sensor fusion unit 31 calculates the fusion reliability and the fusion state. The distance can be obtained from at least one value of the camera sensor 23 and the radar sensor 21. The vehicle speed is detected by the wheel speed sensor 26. The curve R is obtained from the steering angle detected by the steering angle sensor 23 or the road information provided by the navigation system. A steering angle sensor 23 detects the steering angle.

  When the close distance correction flag ON condition is satisfied (Yes in S210), the close distance correction determination unit 32 sets the close distance correction flag to ON (S220).

  Next, the close distance correction determination unit 32 determines whether or not the close distance correction flag OFF condition is satisfied (S230). Thus, the close distance correction determination unit 32 always determines whether the close distance correction flag should be turned on or off.

  When the close condition correction flag OFF condition is satisfied (Yes in S230), the close distance correction determination unit 32 sets the close distance correction flag to OFF (S240).

  If the close distance correction flag is ON (Yes in S250), the correction processing unit 33 performs a close distance correction calculation (S260).

  That is, the correction processing unit 33 employs the shorter of the image distance of the target information C and the radar distance of the target information L as the fusion selection distance. Also, the larger one of the image fusion existence probability and the stationary object existence probability is determined as the image existence probability. Therefore, when a target is present at a close range where the radar sensor is not good like a close-up object still scene, the safer side can be controlled. The fusion selection distance may be used only when the difference between the radar distance and the image distance is less than 1 m. This is because when the difference between the radar distance and the image distance is large, the nearer distance may be less reliable. In addition, in order to determine the collision, the relative speed is acquired from the target information whose distance is selected as the fusion selection distance.

  The collision determination unit 36 calculates the TTC based on the fusion selection distance and the relative speed among the targets determined to possibly come in contact from the lateral position of the target information F, and operates the target device having the smallest TTC. It is determined whether or not 15 is necessary. For example, when the TTC reaches 1 second, the alarm buzzer sounds and when the TTC reaches 0.6 seconds, the automatic brake is activated.

  When the close distance correction flag is not ON (No in S250), the collision determination unit 36 collides based on either the fused target information F output from the sensor fusion unit 31 or the target information L or C. Make a decision. As for the fused target information F, the target information L and the target information C are integrated, so that accurate collision determination is possible. Further, the image fusion existence probability may be used.

  On the other hand, since it is not fusion, the collision determination based on only one of the target information L or C may not be accurate. However, even when only the target information L is detected, the image fusion existence probability is calculated. Therefore, if the image fusion existence probability is equal to or higher than the threshold value, accurate collision determination is possible.

  Further, when only the target information C is detected, if the stationary object existence probability is equal to or greater than the threshold value, the collision is determined based on the target information C. In a close distance passing scene, it is considered that the target enters the detection range from outside the detection range of the camera sensor 23, but the close distance correction flag is turned off depending on the fusion reliability and the steering angle. In this case, since the target only temporarily falls within the detection range of the camera sensor 23, the probability that the stationary object exists exceeds the threshold value is small. Therefore, in a passing scene at a close distance, the collision determination unit 36 hardly determines a collision based on the target information C.

  Further, when the close distance correction flag is OFF, but the target enters the detection range of only the camera sensor 23 from the detection range of the radar sensor 21 and the camera sensor 23, there is a possibility that the stationary object existence probability exceeds the threshold value. high. Therefore, it is possible to make a collision determination with respect to a target detected only by the camera sensor 23. In this case, there is a high possibility that the image fusion existence probability has been calculated in the past for the target, but the image fusion existence probability may exceed the threshold. Therefore, it is possible to make a collision determination based on the image fusion existence probability.

  As described above, the object detection apparatus 100 according to the present embodiment can identify only a scene that is preferably shaken to the safe side by using the close distance correction flag, and can make a collision determination based on the closer radar distance and image distance. It is possible to suppress the uniform extraction of the one closer to the radar distance and the image distance, and the collision determination based only on the image distance when only the target information C is detected.

DESCRIPTION OF SYMBOLS 11 Radar sensor apparatus 12 Camera sensor apparatus 13 Own vehicle information acquisition apparatus 14 System ECU
DESCRIPTION OF SYMBOLS 15 Actuation device 21 Radar sensor 23 Camera sensor 100 Object detection apparatus

Claims (6)

A first sensor for detecting first target information including a first distance to the target and a first lateral position of the target;
A second distance to the target, and second target information including a second lateral position of the target, and a second sensor having a wider detection range of the target than the first sensor;
Own vehicle information acquisition means for acquiring own vehicle information;
For the same target object, the first lateral position and the second lateral position of the same degree is detected more than a predetermined time, the first distance or the second distance is below the threshold, and the vehicle is running straight If you meet the scene specific condition that
Data selection means for selecting a smaller one of the first distance and the second distance as a distance to the target;
When it is determined that the data selection means collides with the target based on the selected distance, driving assistance means for performing driving assistance;
A presence probability calculating means for calculating the presence probability of the target detected by the second sensor so as to increase as the same target is continuously detected;
If the scene specific condition is not satisfied,
The driving support means performs driving support according to a collision determination result with a target based on the second target information if the presence probability calculated by the presence probability calculation means is equal to or greater than a threshold value. An object detection device. Comparing probability calculating means for comparing the first horizontal position and the second horizontal position and calculating a matching probability indicating a possibility that the first target information and the second target information are matched. ,
When the scene specifying condition is satisfied,
The data selecting means selects a larger one of the existence probability and the matching probability as a target existence probability,
The driving support means performs driving support according to a collision determination result with a target based on the distance selected by the data selection means when the presence determination probability is equal to or higher than a threshold value.
The object detection apparatus according to claim 1 . When the second lateral position that is approximately the same as the first lateral position is detected for a predetermined time or more, the second lateral position that is approximately the same as the first lateral position is currently detected, and
Past a predetermined time, if the second lateral position comparable to the first lateral position is detected, or past a predetermined number of times or more, said second lateral position of the same level as the first lateral position The object detection apparatus according to claim 1, wherein the object detection apparatus is a case where the object is detected. The first sensor is a radar sensor and the second sensor is a camera sensor;
Object detection apparatus according to any one of claims 1 to 3, characterized in that. A first sensor for detecting first target information including a first distance to the target and a first lateral position of the target;
A second distance to the target, and second target information including a second lateral position of the target, and a second sensor having a wider detection range of the target than the first sensor;
A host vehicle information acquisition means for acquiring host vehicle information, and a connected information processing apparatus,
For the same target, the second lateral position equivalent to the first lateral position is detected for a predetermined time or more, the first distance or the second distance is equal to or less than a threshold value, and the host vehicle travels straight ahead. If you meet the scene specific condition that
Data selection means for selecting a smaller one of the first distance and the second distance as a distance to the target;
When it is determined that the data selection means collides with the target based on the selected distance, driving assistance means for performing driving assistance;
A presence probability calculating means for calculating the presence probability of the target detected by the second sensor so as to increase as the same target is continuously detected;
If the scene specific condition is not satisfied,
The driving support means performs driving support according to a collision determination result with a target based on the second target information if the presence probability calculated by the presence probability calculation means is equal to or greater than a threshold value. Information processing apparatus. A first sensor for detecting first target information including a first distance to the target and a first lateral position of the target;
A second distance to the target, and second target information including a second lateral position of the target, and a second sensor having a wider detection range of the target than the first sensor;
A host vehicle information acquisition means for acquiring host vehicle information, and an object detection method of a connected apparatus,
For the same target, the second lateral position equivalent to the first lateral position is detected for a predetermined time or more, the first distance or the second distance is equal to or less than a threshold value, and the host vehicle travels straight ahead. If you meet the scene specific condition that
A data selecting means selecting a smaller one of the first distance and the second distance as a distance to the target;
When it is determined that the driving support means collides with the target based on the distance selected by the data selection means;
A presence probability calculating means calculating the presence probability of the target detected by the second sensor so as to increase as the same target is continuously detected;
If the scene specific condition is not satisfied,
The driving support means performs driving support according to a collision determination result with a target based on the second target information if the presence probability calculated by the presence probability calculation means is equal to or greater than a threshold value. An object detection method.

Images