TWI645997B - Obstacle detection credibility evaluation method - Google Patents

Abstract

An obstacle detection reliability evaluation method is implemented by a processing unit, comprising: (A) after receiving a current image, using a first classification method to obtain an indication that the first classification method detects a first current obstacle detection result of the obstacle, and using a second classification method to obtain a second current obstacle detection result indicating the obstacle detected by the second classification method; (B) determining Whether the first and second current obstacle detection results have at least one different obstacle; and (C) determining that the first current obstacle detection result and the second current obstacle detection result are present at least When a different obstacle is obtained, at least one penalty score corresponding to at least one different obstacle is obtained, and a confidence score is subtracted from the at least one penalty score.

Description

Obstacle detection credibility evaluation method

The invention relates to an image data processing, in particular to an obstacle detection reliability evaluation method.

With the popularity of cameras and the development of computer vision, the application of intelligent image surveillance brings safety and convenience to human life, such as the application of advanced identification assistance systems (ADAS) to image cameras. The captured image is detected as an obstacle to warn of dangerous road conditions and, in some cases, to slow or stop the vehicle.

However, the actual road is a relatively complex environment and is susceptible to the intervention of external environmental factors. When the camera is affected by external environmental factors (for example, direct light from a vehicle facing the lane, backlighting, or contamination of the camera lens), The texture and contour of the obstacles in the captured image will also be affected, resulting in inaccurate detection of obstacles. However, the obstacle detection method of the existing ADAS system only outputs the detection result of the obstacle, and does not provide the reliability of the detection result, which will cause the user to mistakenly believe that the detection result of the ADAS system is accurate. Extremely high, and too convinced and rely on the detection results of the ADAS system.

Accordingly, it is an object of the present invention to provide an evaluation method that provides credibility of obstacle detection in a timely manner.

Therefore, the obstacle detection reliability evaluation method of the present invention is implemented by a processing unit electrically connected to a storage unit and an image capturing unit, the storage unit stores a confidence score, and the image capturing unit continuously captures and transmits A series of images are sent to the processing unit, and the obstacle detection reliability evaluation method includes a step (A), a step (B), and a step (C).

In the step (A), after the processing unit receives a current image from the image capturing unit, the processing unit obtains and stores, according to the current image, a first classification method to indicate the first classification method. a first current obstacle detection result of the detected obstacle, and obtaining and storing a second current obstacle indicating the obstacle detected by the second classification method by using a second classification method according to the current image Object detection results.

In the step (B), the processing unit determines whether the first current obstacle detection result and the second current obstacle detection result have at least one different obstacle.

In the step (C), when the processing unit determines that the first current obstacle detection result and the second current obstacle detection result have the at least one different obstacle, the processing unit is based at least according to the at least A different obstacle obtains at least one penalty score corresponding to at least one different obstacle, and subtracts the confidence score from the at least one penalty score to update the confidence score.

The effect of the present invention is that the processing unit determines whether the first current obstacle detection result and the second current obstacle detection result have at least one different obstacle to determine whether the at least one penalty score is obtained. And after obtaining the at least one penalty score, the confidence score is subtracted from the at least one penalty score to update the confidence score.

Referring to Figure 1, a system 100 for implementing one embodiment of the obstacle detection confidence assessment method of the present invention is illustrated. The system 100 includes a storage unit 11 , an image capturing unit 12 , and a processing unit 13 electrically connected to the storage unit 11 and the image capturing unit 12 . The image capturing unit 12 continuously captures and transmits a series of images to the processing unit 13. The storage unit 11 stores a confidence score such as 100 points, a first lookup table of an obstacle pair score, a second lookup table of an obstacle pair score, and a first penalty weight corresponding to a first classification method. And a second penalty weight corresponding to a second classification method. The first lookup table includes a plurality of obstacle distances related to the distance between the actual obstacle and the image capturing unit 12, a plurality of obstacle types related to the type of the obstacle, and a plurality of obstacles respectively corresponding to the obstacles The object distance and the scores of the obstacle categories, Table 1 illustrates the first lookup table. The second lookup table includes a plurality of obstacle positions related to the position of the obstacle in an image, and a plurality of scores respectively corresponding to the obstacle positions, and Table 2 illustrates the second lookup table. Table 1  <TABLE border="1" borderColor="#000000" width="85%"><TBODY><tr><td> Obstacle type obstacle distance</td><td> person</td><td> Car</td></tr><tr><td> 10 meters</td><td> 100 </td><td> 70 </td></tr><tr><td> 30 Ruler</td><td> 70 </td><td> 40 </td></tr><tr><td> 50 meters</td><td> 40 </td><td> 10 </td></tr></TBODY></TABLE> Table 2  <TABLE border="1" borderColor="#000000" width="85%"><TBODY><tr><td> Obstacle position</td><td> Score</td></tr><tr ><td> Center</td><td> 1000 </td></tr><tr><td> Center edge </td><td> 50 </td></tr><tr>< Td> edge</td><td> 0 </td></tr></TBODY></TABLE>

Referring to Figures 1 and 2, an embodiment of how the system 100 performs the obstacle detection confidence assessment method of the present invention will be described. The steps involved in this embodiment are described in detail below.

In step 21, after receiving a current image from the image capturing unit 12, the processing unit 13 obtains and stores an indication of the first classification method by using the first classification method according to the current image. a first current obstacle detection result of the detected obstacle (see FIG. 3), and using the second classification method according to the current image to obtain and store an obstacle indicating the second classification method The second current obstacle detection result (see Figure 4). It should be noted that, in this embodiment, the first classification method obtains the contour texture of the current image by using a histogram of oriented gradient (HOG) and a logarithm weighted pattern (LWP), and The obstacle is detected by a support vector machine (SVM), which is a deep learning convolutional neural network (CNN) to detect obstacles, but not This is limited. Regardless of the first classification method or the second classification method, when detecting obstacles, the two not only detect obstacles but also detect obstacles located at the current image and mark the detected objects. The type of obstacle detected (see Figures 3 and 4), so the first current obstacle detection result and the second current obstacle detection result both contain the detected obstacle and the detected obstacle The location of the obstacle at the current image and the type of obstacle detected.

In step 22, the processing unit 13 determines whether the first current obstacle detection result and the second current obstacle detection result have at least one different obstacle. When the processing unit 13 determines that the first current obstacle detection result and the second current obstacle detection result have the at least one different obstacle, proceed to step 23; and when the processing unit 13 determines the first When a current obstacle detection result and the second current obstacle detection result do not exist the at least one different obstacle, proceed to step 24.

In step 23, the processing unit 13 obtains at least one penalty score corresponding to at least one different obstacle according to the at least one different obstacle, and subtracts the confidence score from the at least one penalty score to update the The confidence score stored by the storage unit 11. In this embodiment, it is assumed that there are N different obstacles, N≧1, and step 23 includes sub-steps 231-236 (see FIG. 5), but not limited thereto.

Referring again to Figures 1 and 5, sub-steps 231-236 are further illustrated.

In sub-step 231, initially, the processing unit 13 determines, for the first different obstacle, whether the first current obstacle detection result has a first different obstacle, that is, i=1.

In sub-step 232, the processing unit 13 determines whether the first current obstacle detection result has an i-th different obstacle. When the processing unit 13 determines that the first current obstacle detection result does not have the i-th different obstacle, step 233 is performed; and when the processing unit 13 determines that the first current obstacle detection result exists When i are different obstacles, proceed to step 234.

In sub-step 233, the processing unit 13 obtains a pair of i-th different obstacles according to the i-th different obstacle, and as the first penalty score of one of the at least one penalty score, and The confidence score is subtracted from the first penalty score to update the confidence score stored by the storage unit 11. In this embodiment, the processing unit 13 first estimates a corresponding i-th obstacle according to the position of the current image by using the i-th different obstacle at a position of the current image. The distance between the actual obstacle and the image capturing unit 12, and then according to the estimated distance between the actual obstacle and the image capturing unit 12 corresponding to the i-th different obstacle, the i-th different obstacle a type, matching the first lookup table, obtaining a score corresponding to the i-th different obstacle, and multiplying the obtained score by the first penalty weight to obtain the first penalty score; however, in other In an embodiment, the processing unit 13 may also obtain another corresponding lookup table related to other obstacle attributes and scores according to other attributes of the i-th different obstacles, such as obstacle positions, etc., to obtain corresponding to the i-th a score of a different obstacle, and multiplying the obtained score by the first penalty weight to obtain the first penalty score; even the processing unit 13 may not need to cooperate with the first lookup table to Different obstacles set as the default score, then the pre-multiplied by the fraction first punishment weights to get the first score of punishment, it is not limited thereto. It is to be noted that since the present invention is not characterized by well-known image distance estimation methods known to those skilled in the art, their details are omitted herein for the sake of brevity.

In sub-step 234, the processing unit 13 obtains a pair of i-th different obstacles according to the i-th different obstacle, and as a second penalty score of one of the at least one penalty score, and The confidence score is subtracted from the second penalty score to update the confidence score stored by the storage unit 11. In this embodiment, the processing unit 13 first estimates the actual obstacle corresponding to the i-th different obstacle by using the image distance estimation method according to the position of the i-th different obstacle located at the current image. The distance from the image capturing unit 12, and then according to the estimated distance of the actual obstacle corresponding to the image capturing unit 12 corresponding to the i-th different obstacle, the type of the i-th different obstacle, Matching the first lookup table, obtaining a score corresponding to the i-th different obstacle, and multiplying the obtained score by the second penalty weight to obtain the second penalty score; however, in other embodiments The processing unit 13 may also obtain another lookup table corresponding to other obstacle attributes and scores according to other attributes of the i-th different obstacles, such as obstacle positions, etc., to obtain corresponding to the i-th different Obscuring the score, multiplying the obtained score by the second penalty weight to obtain the second penalty score; even the processing unit 13 may not need to cooperate with the first lookup table to make the i-th difference Obstacle has been set for the preset score, then the pre-multiplied by the fraction of a second to get that punishment weight fraction of a second punishment, it is not limited thereto.

In sub-step 235 following sub-steps 233 and 234, the processing unit 13 determines whether the i-th different obstacle is the Nth different obstacle, that is, whether i=N. When the processing unit 13 determines that the i-th different obstacle is the Nth different obstacle, proceed to step 24; and when the processing unit 13 determines that the i-th different obstacle is not the Nth Substep 236 is performed for different obstacles.

In sub-step 236, the processing unit 13 sets i to i+1 for the i+1th different obstacle. Thereafter, sub-steps 232-236 are repeated until i=N.

In step 24, the processing unit 13 determines, according to a previous image obtained by the processing unit 13, a first previous obstacle detection result of the obstacle detected by the first classification method, and the a first current obstacle detection result, determining whether the first previous obstacle detection result and the first current obstacle detection result have at least one different obstacle, when the processing unit 13 determines the first front When an obstacle detection result and the first current obstacle detection result have the at least one different obstacle, proceed to step 25; and when the processing unit 13 determines the first previous obstacle detection result and the When the first current obstacle detection result does not exist in the at least one different obstacle, proceed to step 26.

In step 25, the processing unit 13 obtains at least one third penalty score corresponding to at least one different obstacle according to the at least one different obstacle, the second lookup table, and the first penalty weight, and the confidence The degree score is subtracted from the at least one third penalty score to update the confidence score stored by the storage unit 11. In this embodiment, for each different obstacle, the processing unit 13 first matches the second lookup table according to the position of the current image or the previous image by the different obstacle, to obtain a corresponding to the difference. a score of the obstacle, and multiplying the obtained score by the first penalty weight to obtain the third penalty score; however, in other embodiments, the processing unit 13 may also be based on other attributes of the different obstacle For example, an obstacle type or an obstacle distance or the like is matched with another lookup table related to other obstacle attributes and scores to obtain a score corresponding to the different obstacle, and the obtained score is multiplied by the first penalty weight. To obtain the third penalty score; even the processing unit 13 may set each different obstacle to the same preset score without matching the second lookup table, and then multiply the preset score by the first penalty. Weighting to obtain the third penalty score is not limited to this. For example, if the first previous obstacle detection result indicates an obstacle A and an obstacle B, and the first current obstacle detection result indicates the obstacle A and an obstacle C, then The processing unit 13 determines that the first previous obstacle detection result and the first current obstacle detection result have two obstacles different from the obstacle B and the obstacle C.

In step 26, the processing unit 13 determines, according to the previous image, a second previous obstacle detection result of the obstacle detected by the second classification method, and the processing unit 13 a second current obstacle detection result, determining whether the second previous obstacle detection result and the second current obstacle detection result have at least one different obstacle, when the processing unit 13 determines the second front When an obstacle detection result and the second current obstacle detection result have the at least one different obstacle, proceed to step 27; and when the processing unit 13 determines the second previous obstacle detection result and the When the second current obstacle detection result does not exist in the at least one different obstacle, proceed to step 28.

In step 27, the processing unit 13 obtains at least one fourth penalty score corresponding to at least one different obstacle according to the at least one different obstacle, the second lookup table, and the second penalty weight, and the confidence The score is subtracted from the at least one fourth penalty score to update the confidence score stored by the storage unit 11. In this embodiment, for each different obstacle, the processing unit 13 first matches the second lookup table according to the position of the current image or the previous image by the different obstacle, to obtain a corresponding to the difference. a score of the obstacle, and multiplying the obtained score by the second penalty weight to obtain the fourth penalty score; however, in other embodiments, the processing unit 13 may also be based on other attributes of the different obstacle For example, an obstacle type or an obstacle distance or the like is matched with another lookup table related to other obstacle attributes and scores to obtain a score corresponding to the different obstacle, and the obtained score is multiplied by the second penalty weight. To obtain the fourth penalty score; even the processing unit 13 may set each different obstacle to the same preset score without matching the second lookup table, and then multiply the preset score by the second penalty. Weighting to obtain the fourth penalty score is not limited to this. It should be particularly noted that in the present embodiment, steps 22 and 23 are before steps 24 to 27, and in other embodiments, steps 24 to 27 may be before steps 22 and 23, that is, the determination result in step 26. If it is negative, step 22 is performed, and when the result of the determination in step 22 is negative, step 28 is performed.

In step 28, the processing unit 13 outputs the confidence score and then resets the confidence score to 100. It is worth mentioning that, in this embodiment, if the confidence score is negative, the processing unit 13 corrects it to 0 and outputs it again, but is not limited thereto, and the processing unit 13 can also directly output a negative The confidence score of the number.

The third penalty score and the fourth penalty score are different obstacles related to the first previous obstacle detection result and the first current obstacle detection result, and the second previous obstacle detection The third penalty score and the fourth penalty score are related to the position of the different obstacle in the previous image or the current image to determine whether the obstacle is different from the second current obstacle detection result. The shooting range of the image capturing unit 12 is entered and exited at different times for the different obstacles, or the system 100 detects a serious error. And because the first classification method corresponds to the first penalty score and the third penalty score, the second classification method corresponds to the second penalty score and the fourth penalty score, so the first penalty score and the third The penalty score corresponds to the first penalty weight, and the second penalty score corresponds to the second penalty weight, the first penalty weight is, for example, 0.2, and the second penalty weight is, for example, 0.8.

Referring to FIG. 1 , FIG. 6 and FIG. 7 , an embodiment of the obstacle detection reliability evaluation method of the present invention will be described below with reference to an application example. FIG. 6 illustrates the previous image, which includes an obstacle A, An obstacle B, and an obstacle C, FIG. 7 illustrates a first previous obstacle detection result obtained by the processing unit 13 by using the first classification method according to the previous image, the first previous obstacle The detection result indicates that the obstacle A is located in the middle of the previous image and the current image, and the type is a car, which is 12 meters away from the image capturing unit, and the obstacle B is located at the edge of the previous image, and the type is The vehicle is located 30 meters away from the image capturing unit 12, and FIG. 8 illustrates a second preceding obstacle detecting result obtained by the processing unit 13 by using the second sorting method according to the previous image, the second previous one The obstacle detection result indicates that the obstacle A is located in the middle of the previous image and the current image, and the type is a car. The image capturing unit is 12 meters away from the image capturing unit, and the obstacle B is located at the edge of the previous image. For the car, from the image capture 12 30 meters, and the obstacle is located in the C before a current image and the center partial edge image, the type of human, 30 meters from the 12 image capture unit. FIG. 9 illustrates the current image, which includes the obstacle A and the obstacle C. FIG. 10 illustrates a first current obstacle detection result obtained by the processing unit 13 according to the current image by using the first classification method. The first current obstacle detection result indicates that the obstacle A is located in the middle of the current image and the current image, and the type is a car, which is 12 meters away from the image capturing unit. FIG. 11 illustrates that the processing unit 13 according to the The current current image uses the second current obstacle detection result obtained by the second classification method, and the second current obstacle detection result indicates that the obstacle A is located in the middle of the current image, and the type is a car, and the image is taken from the image. The unit 12 is ten meters, and the obstacle C is located at the middle edge of the current image, and the type is human, 30 meters away from the image capturing unit 12.

As shown in step 233, the processing unit 13 indicates the obstacle type and the obstacle distance of the different obstacle (the obstacle C) according to the first current obstacle detection result, and the first lookup table. Obtain a score corresponding to the obstacle C (since the obstacle type is human and the obstacle distance is 30 meters, so the score is 70), and then multiply the obtained score (that is, 70) by the first penalty The weight (i.e., 0.2) is used to obtain the first penalty score (i.e., 0.2 x 70 = 14), and the processing unit 13 updates the confidence score to 100-14 = 86. As shown in step 25, the processing unit 13 obtains a score corresponding to the obstacle B according to the position of the previous image and the second lookup table according to the different obstacle (because of the obstacle B) Located at the edge, so the score is 0), and then the obtained score (ie, 0) is multiplied by the first penalty weight to obtain the third penalty score (ie, 0.2×0=0), the processing unit 13 update the confidence score to 86-0=86. As shown in step 27, the processing unit 13 obtains the first score corresponding to the obstacle B according to the position of the previous image and the second lookup table according to the different obstacle (the obstacle B) (due to the obstacle) B is at the edge, so the score is 0), and the obtained score (ie, 0) is multiplied by the second penalty weight (ie, 0.8) to obtain the fourth penalty score (ie, 0.8×0). =0), the processing unit 13 updates the confidence score to 86-0=86. Finally, in step 28, the confidence score output by the processing unit 13 is 86.

In summary, the obstacle detection reliability evaluation method of the present invention is determined by the processing unit 13 determining whether the first previous obstacle detection result and the first current obstacle detection result are at least one different. Whether the obstacle, the second previous obstacle detection result and the second current obstacle detection result have at least one different obstacle, and the first current obstacle detection result and the second current obstacle detection result Determining whether there is at least one different obstacle to determine whether to obtain the at least one first penalty score, the at least one second penalty score, the at least one third penalty score, and the at least one fourth penalty score, and After the penalty score is obtained, the confidence score is subtracted from the penalty score to update and output the confidence score, thereby providing the reliability of the obstacle detection result to the driving reference, so that the object of the present invention can be achieved.

However, the above is only the embodiment of the present invention, and the scope of the invention is not limited thereto, and all the simple equivalent changes and modifications according to the scope of the patent application and the patent specification of the present invention are still Within the scope of the invention patent.

100‧‧‧ system

11‧‧‧ storage unit

12‧‧‧Image Capture Unit

13‧‧‧Processing unit

21~28‧‧‧Steps

231~236‧‧‧ substeps

A‧‧‧ obstacles

B‧‧‧ obstacles

C‧‧‧ obstacles

Other features and effects of the present invention will be apparent from the following description of the drawings. FIG. 1 is a block diagram illustrating an exemplary method for evaluating the reliability of obstacle detection in the present invention. FIG. 2 is a flow chart illustrating the embodiment of the obstacle detection reliability evaluation method of the present invention; FIG. 3 is a schematic diagram illustrating a first current obstacle detection result; FIG. A schematic diagram illustrating a second current obstacle detection result; FIG. 5 is a flow chart, and FIG. 2 illustrates a sub-step of a step 23 of the embodiment; FIG. 6 is a schematic diagram illustrating a previous image; It is a schematic diagram, with FIG. 6 illustrating a first previous obstacle detection result; FIG. 8 is a schematic diagram, with FIG. 6 illustrating a second previous obstacle detection result; FIG. 9 is a schematic diagram illustrating a current image. FIG. 10 is a schematic diagram showing another first current obstacle detection result in conjunction with FIG. 9; and FIG. 11 is a schematic diagram, and FIG. 9 illustrates another second current obstacle detection result.

Claims (10)

An obstacle detection reliability evaluation method is implemented by a processing unit electrically connected to a storage unit and an image capturing unit, the storage unit stores a confidence score, and the image capturing unit continuously captures and transmits a series of images to The processing unit, the obstacle detection reliability evaluation method includes the following steps: (A) after receiving a current image from the image capturing unit, obtaining and storing an indication according to the current image by using a first classification method a first current obstacle detection result of the obstacle detected by the first classification method, and obtaining and storing, according to the current image, a second classification method, indicating that the second classification method detects a second current obstacle detection result of the obstacle; (B) determining whether the first current obstacle detection result and the second current obstacle detection result have at least one different obstacle; and (C) determining When the first current obstacle detection result and the second current obstacle detection result have the at least one different obstacle, at least according to the at least one different obstacle Obtaining at least a fraction of at least respectively should punish a different obstacles, and minus the confidence score for the punishment of at least a fraction, to update the confidence score. The obstacle detection reliability evaluation method according to claim 1, wherein the step (C) comprises the following substeps: (C-1) determining the first current obstacle detection result and the second current When the obstacle detection result has the at least one different obstacle, determining the first current obstacle for each of the at least one different obstacle Detecting whether the different obstacles are not present; (C-2) when it is determined that the first current obstacle detection result does not exist the different obstacles, at least according to the different obstacles, obtaining at least one a first penalty score of one of the penalty scores, and subtracting the confidence score from the first penalty score to update the confidence score; and (C-3) when determining the first current obstacle detection When the different obstacle exists in the measurement result, at least a second penalty score as one of the at least one penalty score is obtained according to the different obstacle, and the confidence score is subtracted from the second penalty score to Update the confidence score. The obstacle detection reliability evaluation method according to claim 2, wherein the storage unit further stores a first lookup table of the obstacle pair score, a first penalty weight corresponding to the first classification method, and a corresponding The second penalty weight of the second classification method, the first lookup table includes a plurality of obstacle distances related to a distance between an actual obstacle and the image capturing unit, and a plurality of obstacle distances respectively corresponding to the obstacle distances a score, wherein: in the step (A), the first current obstacle detection result and the second current obstacle detection result both include the detected obstacle, and the detected obstacle is located in the The position of the current image; in the step (C-2), the processing unit first estimates the actual obstacle corresponding to the different obstacle and the image capturing unit according to the position of the different obstacle at the current image. a distance, and according to the estimated distance and the first lookup table, obtaining a score corresponding to the different obstacle, and multiplying the obtained score by the first penalty weight to obtain the first penalty score; and In the step (C-3), the processing unit first estimates the distance between the actual obstacle corresponding to the different obstacle and the image capturing unit according to the position of the different obstacle at the current image, and according to The estimated distance and the first lookup table obtain a score corresponding to the different obstacle, and multiply the obtained score by the second penalty weight to obtain the second penalty score. The obstacle detection reliability evaluation method according to claim 2, wherein the storage unit further stores a first lookup table of the obstacle pair score, a first penalty weight corresponding to the first classification method, and a corresponding The second penalty weight of the second classification method, the first lookup table includes a plurality of obstacle categories, and a plurality of scores respectively corresponding to the categories, wherein: in step (A), the first current obstacle The object detection result and the second current obstacle detection result both include the detected obstacle and the type of the detected obstacle; in step (C-2), the processing unit is based on the Different types of obstacles and the first lookup table, obtaining a score corresponding to the different obstacles, and multiplying the obtained score by the first penalty weight to obtain the first penalty score; and at the step ( In C-3), the processing unit obtains a score corresponding to the different obstacle according to the type of the different obstacle and the first lookup table, and multiplies the obtained score by the second penalty weight. Obtain the second penalty score. The method for evaluating the obstacle detection reliability described in claim 1 further comprises the following steps after the step (A): (D) determining, according to a first image obtained by the first image, a first previous obstacle detection result of the obstacle detected by the first classification method and the first current obstacle detection result Whether the first obstacle detection result and the first current obstacle detection result have at least one different obstacle; (E) determining the first previous obstacle detection result and the first current obstacle Detecting that at least one different obstacle exists, at least one third penalty score corresponding to at least one different obstacle is obtained according to the at least one different obstacle, and the confidence score is subtracted from the at least one The third penalty score is to update the confidence score. The obstacle detection reliability evaluation method according to claim 5, wherein the storage unit further stores a second lookup table of the obstacle pair score, wherein the second lookup table includes a plurality of positions related to the obstacle located at an image. a position of the obstacle, and a plurality of scores respectively corresponding to the positions of the obstacles, wherein: in the step (A), the first current obstacle detection result and the second current obstacle detection result both include Detecting the obstacle and the detected obstacle is located at the current image; and in step (E), the processing unit is located in the current image or the previous image according to the different obstacle The location and the second lookup table obtain a score corresponding to the different obstacle, and then obtain the third penalty score according to the obtained score. The method for evaluating the obstacle detection reliability according to claim 1, further comprising the following steps after the step (A): (F) indicating the second classification party according to an indication obtained by the previous image. Determining the second previous obstacle detection result and the second current obstacle detection result by the second previous obstacle detection result of the obstacle detected by the method and the second current obstacle detection result Whether there is at least one different obstacle; (G) at least according to the at least one when it is determined that the second previous obstacle detection result and the second current obstacle detection result have the at least one different obstacle The different obstacles obtain at least one fourth penalty score that should be at least one different obstacle, and subtract the confidence score from the at least one fourth penalty score to update the confidence score. The obstacle detection reliability evaluation method according to claim 7, wherein the storage unit further stores a second lookup table of the obstacle pair score, wherein the second lookup table includes a plurality of positions related to the obstacle located at an image. a position of the obstacle, and a plurality of scores respectively corresponding to the positions of the obstacles, wherein: in the step (A), the first current obstacle detection result and the second current obstacle detection result both include The detected obstacle and the detected obstacle are located at the current image; and in step (G), the processing unit is located in the current image or the previous image according to the different obstacle The location and the second lookup table obtain a score corresponding to the different obstacle, and then obtain the fourth penalty score according to the obtained score. The obstacle detection reliability evaluation method according to claim 1, wherein in the step (A), the first classification method detects obstacles by contour texture, and the second classification method is deep learning detection. Measure obstacles. The obstacle detection reliability evaluation method according to claim 9, wherein In the step (A), the first classification method obtains the current image contour texture by a gradient direction histogram and a logarithmic weighting mode, and detects an obstacle by a support vector machine, and the second classification method is a convolution of deep learning. The neural network detects obstacles.