TWI531499B - Anti-collision warning method and device for tracking moving object - Google Patents

Description

Anti-collision warning method and device capable of tracking moving object

The invention relates to a collision avoidance warning method and a device thereof, in particular to a collision avoidance warning method and a device thereof for tracking a moving object.

According to the vehicle, as a carrier and transportation vehicle, the vehicle has already played an important and indispensable role in human life. However, in the narrow and dense living environment, traffic accidents are repeated, and there are many reasons for traffic accidents. It can be roughly divided into natural environmental factors and human factors. In order to effectively reduce the incidence of traffic accidents, relevant prevention The driving safety warning technology of the vehicle in collision with other vehicles or pedestrians has been developed successively for the owner to select to improve driving safety.

For products that continue to be driving safety warnings, the most common use of GPS positioning systems to detect obstacles and their relative distance from existing vehicles, however, GPS is limited to environmental factors, such as the current vehicle to the sheltered area The obstacle can not be detected, and the utility is limited for the driver; or the distance sensor, the image sensor, etc., the distance sensor mainly assists the obstacle in one direction, and the image sensor The visual aid applied to the wide area can effectively help the driver to grasp the relative distance between the current vehicle dynamics and the obstacles to reduce the occurrence of collision accidents.

Furthermore, in order to improve the accuracy of estimating the relative distance between current vehicle dynamics and obstacles, a Kalman filter algorithm is proposed to estimate the relative distance between the current vehicle and the obstacle and the collision time prediction method. However, this method The algorithm is only an object estimation suitable for linear movement, and cannot be applied to predict the possibility of collision from any direction toward the driver, so the practical benefit is limited.

In view of the above, the present invention provides a collision avoidance warning method and a device thereof for tracking a moving object in order to effectively overcome the above problems.

The main object of the present invention is to provide a collision avoidance warning method and a device thereof for tracking a moving object, which can classify the type of the obstacle according to the characteristics of the length and width of the obstacle, thereby tracking the dynamics of the obstacle to improve the estimated collision time. Accuracy.

A secondary object of the present invention is to provide a collision avoidance warning method and apparatus for tracking a moving object, which uses an extended Kalman filter algorithm to track the movement of obstacles and filter out noise on the sensing, which is suitable for non- Linearly moving obstacles can effectively reduce the jitter of the collision time value estimated by the Kalman filter algorithm, and improve the reliability of use.

In order to achieve the above object, the present invention provides a collision avoidance warning method for tracking a moving object, which is suitable for use in a vehicle. The anti-collision warning method comprises the following steps: after capturing a plurality of consecutive images, identifying the continuous images At least one obstacle, and obtain geometric feature parameters and image pixel feature parameters of the obstacle length and width, and use a binary tree classifier to quickly classify the obstacle type. According to the type of the obstacle, at least one dynamic obstacle is found. Since the dynamic obstacle may be linearly moved or nonlinearly moved, the continuous relative position of the dynamic obstacle and the vehicle is first detected, and one of the vehicles is first estimated. Collision area. Then, according to the continuous relative position and an extended Kalman filter algorithm, the moving speed, the moving direction and the current position of the dynamic obstacle are estimated, thereby obtaining one of the dynamic obstacles and the second collision area; and according to the first collision area and The second collision area estimates a collision point, and determines whether the first collision area and the second collision area at least partially overlap, and if so, estimates a collision time, and outputs a warning signal to promptly warn the driver; if not, The step of capturing a plurality of consecutive images is repeated.

The present invention provides another anti-collision warning device for tracking a moving object, which is disposed on a vehicle. The anti-collision warning device includes at least two image capturing units, a body signal sensing unit, an image processing module, and a central processing unit. And a warning unit. At least two image capturing unit images capture a plurality of images in the front region of 180 degrees, respectively capturing the distant field of view image and the near field of view image, thereby expanding the detection field of view. The image processing module is electrically connected to the second image capturing unit, and identifies at least one obstacle in the images and the relative position with the vehicle, and obtains geometric feature parameters and image pixel characteristic parameters of the obstacle length and width, and uses the second A meta-tree classifier quickly classifies the type of obstacle and at least one of its dynamic obstacles. The body signal sensing unit senses the dynamic signal of the vehicle. The central processing unit is electrically connected to the body signal sensing unit and the image processing module. The central processor calculates the relative position of the dynamic obstacle and the vehicle according to the dynamic signal of the vehicle and the dynamic obstacle, and estimates one of the vehicles. a collision region; and using an extended Kalman filter algorithm to obtain a second collision region of one of the dynamic obstacles, and estimating a collision point according to the first collision region and the second collision region; when the first collision region and the first collision region When the two collision regions are at least partially overlapped, a collision time is estimated, and a control signal is output, and the warning unit is electrically connected to the central processing unit, and receives the control signal to output a warning signal to promptly warn the driver.

The purpose, technical content, features and effects achieved by the present invention will be more readily understood by the detailed description of the embodiments.

In order to provide a more accurate warning signal of the collision point and collision time to the driver, the driver can instantly grasp the relative position and movement between the current vehicle and the obstacle to avoid the occurrence of a collision accident. Anti-collision warning method for reliable tracking of moving objects and Its device is used to achieve the purpose of immediate collision prevention.

As shown in Fig. 1, it is a block diagram of the circuit of the present invention. The anti-collision warning device is disposed on a vehicle. The anti-collision warning device 10 includes at least two image capturing units 12, a body signal sensing unit 14, an image processing module 16, a central processing unit 18, and a warning unit 20. At least two image capturing units 12 capture a plurality of consecutive images of 180 degrees in the front region, respectively capturing the distant field of view image and the near field of view image, thereby expanding the detection field of view. The body signal sensing unit 14 senses the dynamic signal of the vehicle. The image processing module 16 is electrically connected to the image capturing unit 12, and the image processing module 16 identifies the distance vision image and the near field of view image and the elevation angle of the second image capturing unit 12 in the continuous image to identify at least An obstacle, and calculating a continuous relative position of the obstacle and the vehicle, and using a binary tree classifier to quickly classify the obstacle type and at least according to the geometric characteristic parameter of the obstacle length and width and the image pixel characteristic parameter A dynamic obstacle. The central processing unit 18 is electrically connected to the body signal sensing unit 14, the image processing module 16, and the warning unit 20. The body signal sensing unit 14 senses dynamic signals of the vehicle, such as driving directions, driving speed, and the like, and central processing. The device 18 calculates a continuous relative position of the dynamic obstacle and the vehicle according to the dynamic signal and the dynamic obstacle in the running of the vehicle, and estimates one of the first collision regions of the vehicle, and uses the extended Kalman filter algorithm to obtain dynamics. a second collision area of the obstacle; estimating a collision point according to the first collision area and the second collision area, and estimating a collision time when the first collision area and the second collision area at least partially overlap, and outputting A control signal, after receiving the control signal, the warning unit 20 outputs a warning signal correspondingly, so that the driver can be alerted to avoid collision.

In addition to using the second image capturing unit 12 to capture the obstacle image, the at least one ranging sensor 22 can be matched with the vehicle, and electrically connected to the central processing unit 18 for ranging sensing. The device 22 cooperates with the second image capturing unit 12 to instantly detect the relative position of the dynamic obstacle and the vehicle. The ranging sensor 22 is a radar sensor, a light radar sensor, an ultrasonic sensor, or an infrared sensor.

In order to further understand the anti-collision warning method of the present invention, please refer to FIG. 1 and FIG. 2 together, and FIG. 2 is a flow chart of the steps of the present invention. First, in step S10, the at least two image capturing units 12 respectively capture a plurality of consecutive images of the distant view image and the close-view image to capture a wide range of view range images. In step S12, the image processing module 16 identifies at least one obstacle in the continuous images, and uses a feature algorithm to obtain geometric feature parameters and image pixel feature parameters of the obstacle length and width, which are assumed to be forward. Obstacle, you can use the binary tree classifier to quickly classify the types of obstacles. Please also cooperate with Figure 3, the feature algorithm contains the following formula:

Where f is the focal length of the image capturing unit (such as the distance from the image plane to the center of the lens), and x and y are the pixel positions of the image plane, that is, the origin of the image plane, where the origin is the center point of the image plane, For example, 720 ^* 480 images, then (x, y) is the (360, 240) image plane center point. The X, Y, and Z systems are the world coordinates of the obstacle relative to the image capturing unit; h is the height of the image capturing unit.

After the image processing module 16 is calculated, the geometric feature parameters and the image pixel characteristic parameters of the obstacle can be obtained, and the types of obstacles can be classified according to the classification, for example, can be divided into pedestrians, motorcycles, buses, small Types of passenger cars or road environments. Among them, the Histogram of oriented gradient (HOG) or the Haar Feature can be used to identify Obstacle features, and with the support vector machine (SVM) or classifier of the Artificial Neural Network (ANN) to accurately classify pedestrians or motorcycles (locomotives and bicycles) Or, with the image width and height geometry, and with LDA feature space conversion to classify large obstacles such as buses or passenger cars. Next, in step S14, according to the type of the obstacle and its continuous moving image, at least one dynamic obstacle is found, which is the moving object of interest to be tracked dynamically according to the present invention.

In step S16, a continuous relative position of the dynamic obstacle (here, a person as an example) to the vehicle is detected, and one of the first collision areas of the vehicle is estimated, wherein the image processing module 16 can identify the continuous After the image, the continuous relative position of the dynamic obstacle and the vehicle is detected, or the image capturing unit 12 and the distance measuring sensor 22 are integrated to detect the continuous relative position of the dynamic obstacle and the vehicle, no matter which one is used. In the manner of detection, the first collision area of the vehicle can be estimated based on the continuous relative position of the dynamic obstacle and the vehicle.

Since the dynamic obstacle 24 is not limited to linear movement, in order to more accurately estimate the dynamics of the dynamic obstacle, in step S18, the central processing unit 18 according to the continuous relative position, such as detecting the relative distance and relative of the dynamic obstacle to the vehicle. Angle, and an extended Kalman filter algorithm to estimate the moving speed, moving direction and current position of the dynamic obstacle, thereby obtaining a second collision area of one of the dynamic obstacles. Among them, the extended Kalman filter algorithm contains the following formula:

Where xp _i is the x-axis position of the dynamic obstacle, yp _i is the y-axis position of the dynamic obstacle, v _i is the speed of the dynamic obstacle, φ _i is the traveling direction of the dynamic obstacle, and Δ t is the input dynamic obstacle The sampling time of the continuous relative position with the vehicle, A is a state transition model of the dynamic obstacle, Estimating the vector for the previous step state, For the current observation vector.

Next, in step S20, a collision point is estimated according to the first collision area and the second collision area. As shown in FIG. 4, the dynamic obstacle 24 position (A), the vehicle 26 position (B), and the collision point (C) A triangular geometric relationship is formed, wherein the known parameters are: vehicle 26 heading angle (H _B ), dynamic obstacle 24 heading angle (H _A ), dynamic obstacle 24 angle relative to vehicle 26 (H _AB ), vehicle 26 with respect to the angle of the dynamic obstacle 24 (H _BA ), the relative linear distance (D) of the vehicle 26 from the dynamic obstacle 24, the two internal angles ∠ A and ∠ B and the collision angle ∠ C can be calculated from known parameters. ; according to the following sine theorem: After the operation, the distance of the vehicle 26 position (B) from the collision point (C) and the distance of the dynamic obstacle 24 (A) from the collision point (C) can be obtained. Next, in step S22, the central processing unit 18 determines whether the first collision area and the second collision area at least partially overlap, and if not, repeats step S10; if so, it is highly probable that a collision occurs within a few seconds, so execution is performed. In the next step S24, a collision time is estimated, and a warning signal is output, so that the driver can immediately grasp the relative position and movement between the current vehicle 26 and the dynamic obstacle 24 to avoid a collision accident. Among them, the estimated collision time mode can be referred to Fig. 4 at the same time. The estimated collision time can be divided into longitudinal collision time and lateral collision time. The longitudinal collision time ( t _ADM ) of the dynamic obstacle 24 relative to the collision point (C) is based on the following formula. Obtained: ; and e _A = α. Obj _w ; where V _A is the moving speed of the dynamic obstacle, ADM is the distance between the dynamic obstacle position and the collision point, e _A is the estimated error value of the dynamic obstacle width, and α is the second of these continuous images. The error coefficient of the image capturing unit, obj _w is the image capturing unit to identify the width of the dynamic obstacle.

The longitudinal collision time ( t _BDM ) of the vehicle 26 relative to the point of impact (C) is obtained according to the following formula:

Where V _B is the vehicle speed of the vehicle and BDM is the distance between the vehicle position and the collision point.

The longitudinal collision time is based on the moving speed of the dynamic obstacle and the distance ADM from the collision point (C) to find the required time t _{ADM to} reach the collision point (C), and the vehicle speed and the collision point of the vehicle (C The distance BDM is used to find the time t _BDM required to reach the collision point (C); when the time of t _ADM and t _BDM overlaps, it is the longitudinal collision time of the vehicle and the dynamic obstacle.

The so-called lateral collision time judgment means that the vehicle 26 and the dynamic obstacle 24 continue to advance, possibly before reaching the collision point, because the dynamic obstacle has a large length and width (for example, a large container truck or a junction vehicle), A lateral collision accident occurs, so the lateral collision time ( t _LSM ) of the vehicle 26 with the dynamic obstacle 24 is additionally considered, which is obtained according to the following formula:

Where D is the relative linear distance between the vehicle and the dynamic obstacle. According to the first collision area, the second collision area and the collision point, the two internal angles ∠ A and ∠ B and the collision angle ∠ C and β can be obtained as detection dynamics. The error coefficient of the continuous relative position of the obstacle to the vehicle. When t _{LSM is} less than a preset value, it is the lateral collision time of the vehicle and the dynamic obstacle.

Therefore, after the collision point and the collision time are obtained, the control signal is output to the warning unit 20 through the central processing unit 18, and the warning unit 20 outputs a warning signal corresponding to the driver. The warning unit 20 is a display, and can display an overlapping screen of the first collision area and the second collision area. And the point in time of the collision, or the display integrated with the voice system, simultaneously displaying and voice-informing the driver of the relevant collision information.

Please refer to FIG. 1 and FIG. 5 at the same time. FIG. 5 is a flow chart of steps for detecting a dynamic obstacle according to the present invention. When the driving environment is in a dim state, for example, the evening sun obliquely makes the night light overexposed or the brightness is insufficient, or the rainy day causes the influence characteristic to be blurred, etc., in order to accurately grasp the running road condition, first, step S26 is performed, and at least two images are captured. A continuous image of 180 degrees in the front area of the unit 12 is further recognized by the image processing module 16 in step S28. The central processing unit 18 determines whether the image capturing unit 12 is in a failure mode according to the sharpness of the continuous images. That is, whether the continuous images have clear images of at least one obstacle; if the mode is normal, step S30 is performed, and at least one obstacle is recognized by the image processing module 16, and the geometric characteristic parameters of the obstacle length and width are obtained. And the image pixel feature parameter; if not, the image capturing unit 12 is in a failure mode, and in this case, step S32 is performed, and at least one obstacle is detected by the ranging sensor 22, such as a light radar sensor (Lidar). And obtain the geometric characteristic parameters of the obstacle length and width and the image pixel characteristic parameters, and use the binary tree classifier to classify the obstacles. Further, in step S34, the distance measuring sensor 22 obtains a continuous relative distance between the traveling vehicle and the obstacle, thereby finding at least one dynamic obstacle, which is also interested in tracking dynamics of the present invention. Move the object. Therefore, in addition to using the image capturing unit 12 to obtain an image of an obstacle, the present invention also considers an image blurring problem caused by environmental factors, so the central processing unit 18 is used to control the ranging sensor 22 to detect obstacles. The position, such that the image capturing unit 12 and the ranging sensor 22 are matched with each other, the present invention can effectively improve driving safety.

In summary, the present invention can classify the type and length and width information according to the characteristics of the obstacle, thereby tracking the dynamics of the obstacle, so as to improve the accuracy of estimating the collision time, and can indeed improve the existing The technology can only identify dynamic or static obstacles, while ignoring the size of the current vehicle and obstacles, the shortcomings of estimating the actual value of the collision point and the collision time still have large errors.

Furthermore, although the Kalman filter algorithm has been used to track the motion of obstacles, it is limited to the estimation of linear moving objects. However, most of the dynamic obstacles are nonlinear movements. Therefore, the present invention uses an extended Kalman filter algorithm to track the movement of obstacles, which can cover obstacles moving linearly or nonlinearly, and can filter out the sensing. The noise can effectively reduce the jitter of the collision time value estimated by the Kalman filter algorithm, and improve the reliability of the use.

The above is only the preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Therefore, any changes or modifications of the features and spirits of the present invention should be included in the scope of the present invention.

10‧‧‧Anti-collision warning device

12‧‧‧Image capture unit

14‧‧‧ Body signal sensing unit

16‧‧‧Image Processing Module

18‧‧‧Central processor

20‧‧‧Warning unit

22‧‧‧Ranging sensor

24‧‧‧Dynamic obstacles

26‧‧‧ Vehicles

Figure 1 is a block diagram of the circuit of the present invention.

Figure 2 is a flow chart showing the steps of the present invention.

Figure 3 is a schematic view showing the geometrical features of the obstacle detection in the present invention.

Figure 4 is a schematic diagram of the predicted collision area and time of the present invention.

Figure 5 is a flow chart showing the steps of detecting a dynamic obstacle in the present invention.

Claims (9)

An anti-collision warning method for tracking a moving object, which is applicable to an axle, comprising the steps of: capturing a plurality of consecutive images; identifying at least one obstacle in the continuous images, and obtaining geometric features of the obstacle length and width Parameter and image pixel feature parameters, using a binary tree classifier to classify the type of the obstacle, wherein the step of identifying the obstacle in the images uses a feature algorithm to obtain the geometry of the obstacle length and width Characteristic parameters and image pixel feature parameters, the feature algorithm: Wherein, Y is the Y-axis direction of the image capturing unit, w is the Y-axis downward inclination, h is the erection height of the image capturing unit; and at least one dynamic obstacle is found according to the type of the obstacle; detecting the a continuous relative position of the dynamic obstacle to the vehicle, and estimating a first collision area of the vehicle; estimating the moving speed and moving direction of the dynamic obstacle according to the continuous relative position and an extended Kalman filter algorithm And a current position, according to which a second collision area of the dynamic obstacle is obtained; and a collision point is estimated according to the first collision area and the second collision area, and the first collision area and the second collision are determined. Whether the areas overlap at least partially, and if so, a collision time is estimated and a warning signal is output, and if not, the first step is repeated. The collision avoidance warning method for tracking a movable object according to claim 1, wherein the obstacle type can be classified into a pedestrian, a motorcycle, a bus, a passenger car, or a road environment. The method for detecting a collision avoidance of a movable object according to claim 1, wherein in the step of detecting a relative position of the dynamic obstacle and the vehicle, the dynamic obstacle is detected by the at least one sensor and the The relative position of the relative distance and relative angle of the vehicle. The anti-collision warning method for tracking a movable object according to claim 1, wherein the extended Kalman filter algorithm comprises the following formula: Where xp _{i is} the x-axis position of the dynamic obstacle, yp _{i is} the y-axis position of the dynamic obstacle, v _{i is} the relative speed of the dynamic obstacle and the vehicle, and φ _{i is} the dynamic obstacle and the vehicle opposite direction, △ t is the sampling time and the continuous input of the moving obstacle relative position of the vehicle, a transformation model for the state of the dynamic obstacle, Estimating the vector for the previous step state, For the current observation vector. The anti-collision warning method for tracking a movable object according to claim 1, wherein in the step of estimating the collision time, the collision time is divided into a longitudinal collision time and a lateral collision time, wherein: the dynamic obstacle is opposite to the collision point The longitudinal collision time ( t _ADM ) is obtained according to the following formula: e _A = α . Obj _w ; where V _{A is} the moving speed of the dynamic obstacle, ADM is the distance between the dynamic obstacle position and the collision point, e _A is the estimated error value of the dynamic obstacle width, and α is the value The error coefficient of at least two image capturing units of the continuous image, obj _{w is} the image capturing unit identifying the width of the dynamic obstacle; the longitudinal collision time ( t _BDM ) of the vehicle relative to the collision point is obtained according to the following formula : Wherein, V _{B is} the vehicle speed of the vehicle, and BDM is the distance between the vehicle position and the collision point; when the time of the t _ADM and the t _BDM overlap, the longitudinal collision time of the vehicle and the dynamic obstacle; The lateral collision time ( t _LSM ) of the vehicle and the dynamic obstacle is obtained according to the following formula: Where D is the relative linear distance between the vehicle and the dynamic obstacle, and according to the first collision area, the second collision area and the collision point, two internal angles ∠ A and ∠ B and a collision angle ∠ C can be obtained. β is an error coefficient for detecting the continuous relative position of the dynamic obstacle and the vehicle; when t _{LSM is} less than a preset value, it is the lateral collision time of the vehicle and the dynamic obstacle. An anti-collision warning device capable of tracking a moving object is disposed on a vehicle, the anti-collision warning device comprises: at least two image capturing units, which are continuous images of 180 degrees in front of a plurality of front regions; and a body signal sensing unit Sensing the dynamic signal of the vehicle; an image processing module electrically connecting the two image capturing units, identifying at least one obstacle in the continuous images and its continuous relative position with the vehicle, and obtaining the The geometric feature parameters of the obstacle length and width and the image pixel feature parameters are determined by a binary tree classifier to classify the obstacle type and at least one dynamic obstacle. The image processing module uses the following feature algorithm to obtain The geometric characteristic parameter of the obstacle length and width and the image pixel characteristic parameter, the feature algorithm: Wherein, Y is the Y-axis direction of the image capturing unit, w is the Y-axis downward inclination, h is the erection height of the image capturing unit; a central processor electrically connecting the body signal sensing unit and the An image processing module, the central processor calculates a continuous relative position of the dynamic obstacle and the vehicle according to the dynamic signal and the dynamic obstacle, thereby estimating a first collision area of the vehicle, and using the extended type a Kalman filter algorithm to obtain a second collision region of the dynamic obstacle, and estimating a collision point according to the first collision region and the second collision region, when the first collision region and the second collision region are at least When partially overlapping, a collision time is estimated, and a control signal is output; and a warning unit is electrically connected to the central processing unit to receive the control signal to output a warning signal. The anti-collision warning device of the trackable moving object according to claim 6, wherein the warning unit is a display, and the first collision area and the second collision area are displayed, and the collision point and the collision time are displayed. The collision time is divided into a longitudinal collision time and a lateral collision time, and the longitudinal collision time ( t _ADM ) of the dynamic obstacle relative to the collision point is obtained according to the following formula: e _A = α . Obj _w ; where V _{A is} the moving speed of the dynamic obstacle, ADM is the distance between the dynamic obstacle position and the collision point, e _A is the estimated error value of the dynamic obstacle width, and α is the value The error coefficient of the at least two image capturing units of the continuous image, obj _{w is} the image capturing unit identifying the width of the dynamic obstacle; the longitudinal collision time ( t _BDM ) of the vehicle relative to the collision point is obtained according to the following formula Get: Wherein, V _{B is} the vehicle speed of the vehicle, and BDM is the distance between the vehicle position and the collision point; when the time of the t _ADM and the t _BDM overlap, the longitudinal collision time of the vehicle and the dynamic obstacle; The lateral collision time ( t _LSM ) of the vehicle and the dynamic obstacle is obtained according to the following formula: Where D is the relative linear distance between the vehicle and the dynamic obstacle, and according to the first collision area, the second collision area and the collision point, two internal angles ∠ A and ∠ B and a collision angle ∠ C can be obtained. β is an error coefficient for detecting the continuous relative position of the dynamic obstacle and the vehicle; when the t _{LSM is} less than a preset value, the lateral collision time of the vehicle and the dynamic obstacle is. The anti-collision warning device for tracking a moving object according to claim 6, wherein the two image capturing units respectively capture a distant view image and a close-up view image, according to the distance view image and the near field of view The obstacle in the image and the elevation angle of the two image capturing unit to calculate the relative position of the vehicle and the obstacle. The anti-collision warning device of the trackable moving object according to claim 6, further comprising at least one ranging sensor electrically connected to the central processing unit, wherein the ranging sensor is coupled to the two image capturing unit, To detect the relative position of the dynamic obstacle and the vehicle, the ranging sensor is a radar sensor, a light radar sensor, an ultrasonic sensor or an infrared sensor.