Life body radar detection system and its judgment method, feature database establishment method

Abstract

The present invention provides a radar detection system for life forms, a judgment method and a method for establishing a feature database. The system comprises an image analysis unit, a spectrum analysis unit and an operation processing unit. The operation processing unit comprises a feature extraction unit and a matching judgment unit. The image analysis unit and the spectrum analysis unit receive signals detected by radar, and perform signal processing on the signals to form image signals and generate spectrum signals. The feature extraction unit receives the image signals and the spectrum signals to perform feature extraction to obtain the features to be detected of the object to be detected. The matching judgment unit compares the features to be detected to accurately judge whether the object to be detected is a life form. If it is judged to be a life form, it is classified as a control feature of a life form and established in a feature database, so as to accurately judge whether the object to be detected is a life form.

Description

Technical Field

[0001] The present invention relates to a life form detection technology, and more particularly to a life form radar detection system and a judgment method and a feature database establishment method thereof. Background Art

[0002] Advanced Driver Assistance Systems (ADAS) are the main research and development goals of all relevant companies. Regardless of any stage of assisted driving or autonomous driving technology, the most important thing is how to effectively judge emergency situations and take corresponding actions. Among them, the most important thing is how to correctly judge the life forms, such as pedestrians, encountered in front of the vehicle during the driving situation, so that emergency response can be made when there is a risk of collision. However, during the driving process, the time to judge pedestrians and deal with them in real time is very short, and it can only rely on system judgment. Therefore, a more accurate and fast judgment mode is required to meet the driving safety requirements of the advanced driver assistance system. Therefore, how to quickly and correctly judge the presence of life forms such as pedestrians or animals in front during driving, and then make timely braking or evasive response actions to avoid threats to the life safety of life forms, is a very important link in the development of advanced driver assistance system technology.

[0003] According to the existing obstacle detection technology, such as radar detection technology, it is difficult to determine whether the obstacle is a living thing, especially a pedestrian, because the radio wave information it generates can only detect the position and distance of the obstacle, and cannot show the image of a living thing. If the judgment method of image reading is used, although it can determine that the obstacle is a living thing, it has the problem of not being able to accurately determine the moving speed and distance of the living thing.

[0004] As mentioned above, the US patent No. US10354160B2 invented a pedestrian detection device and method, which is mainly a pedestrian detection system that uses a camera to read images and integrates radar detection (Camera-Radar Fusion). In addition to detecting the location and distance of obstacles, it can also determine whether the obstacles are pedestrians through the read image display. However, the image of pedestrians read by the camera is easily disturbed by light. For example, in cloudy, rainy or dark environments, the image will be dim and blurred due to insufficient light, or the image will be overexposed when the light is too bright. These situations will make the pedestrian image unclear, leading to image misjudgment, which seriously threatens the life safety of pedestrians. Summary of the invention

[0005] To solve the above problems, the present invention provides a radar detection system for detecting life forms, a judgment method thereof, and a feature database establishment method, which uses image analysis and spectrum analysis as the main detection technologies to determine whether the object to be detected is a life form.

[0006] An embodiment of the present invention provides a radar detection system for detecting life forms and a method for determining life forms. The system includes an image analysis unit, a spectrum analysis unit, and an operation processing unit. The operation processing unit is electrically connected to the image analysis unit and the spectrum analysis unit. The operation processing unit includes a feature extraction unit and a matching determination unit. The image analysis unit receives a signal measured by a radar, and performs signal processing on the signal to generate a plurality of sampling points to form an image signal. The spectrum analysis unit receives the signal and performs signal processing on the signal to generate a spectrum signal. The feature extraction unit receives the image signal and the spectrum signal to perform feature extraction to obtain a feature to be detected of an object to be detected. The matching determination unit compares the feature to be detected to determine whether the object to be detected is a life form.

[0007] Thus, when an object to be detected is moving, the feature extraction unit receives image signals and spectrum signals to extract features and obtain the features to be detected, and the matching judgment unit compares the features to be detected, so as to combine the above two phenomena and complementarily judge the real movement mode of the object to be detected, thereby solving the problem of image misjudgment caused by light interference when existing cameras read images, and can achieve the effect of accurately judging whether the object to be detected is a living organism.

[0008] Another embodiment of the present invention provides a method for establishing a feature database of a life form detection radar detection system, which collects a radar signal from a life form, including receiving the signal with an image analysis unit and performing signal processing on the signal to generate multiple sampling points to form an image sample signal. On the other hand, the signal is also received with a spectrum analysis unit and processed to generate a spectrum sample signal. Then, a feature extraction unit extracts features from the image sample signal and the spectrum sample signal to obtain a comparison feature, which is established in a feature database and classified as a comparison feature of the life form.

[0009] Thus, as the feature database is established, the control features classified as living organisms can be used as the standard for the matching judgment unit to compare the features to be detected, so that there is a basis for judging whether the object to be detected is a living organism. Furthermore, the feature database can continuously establish different control features that can be classified as living organisms, and by continuously expanding different control features in the feature database, the life body detection radar detection system can achieve artificial intelligence deep learning when judging whether the object to be detected is a living organism, so that the matching judgment unit can have sufficient comparison samples as reference when comparing the features to be detected, which can ensure that the object to be detected is judged to be a living organism with a high degree of accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0011] Figure 1 4 is a system block diagram of a radar detection system for detecting life forms according to an embodiment of the present invention.

[0012] Figure 2 4 is another system block diagram of the radar detection system for life body detection according to an embodiment of the present invention.

[0013] Figure 3 It is a flow chart of a method for determining a living being according to an embodiment of the present invention.

[0014] Figure 4 It is a flow chart of a method for establishing a feature database according to an embodiment of the present invention.

[0015] Figure 5 This is a sampling point distribution diagram of the image signal measured when the object to be detected is a tree according to an embodiment of the present invention.

[0016] Figure 6 It is a waveform diagram of a frequency spectrum signal measured when the object to be detected is a tree according to an embodiment of the present invention.

[0017] Figure 7 It is a sampling point distribution diagram of the image signal measured when the object to be detected is a dog in the embodiment of the present invention.

[0018] Figure 8 It is a waveform diagram of the frequency spectrum signal measured when the object to be detected is a dog according to the embodiment of the present invention.

[0019] Fig. 9 It is a sampling point distribution diagram of the image signal measured when the object to be detected is a pedestrian in the embodiment of the present invention.

[0020] Fig.10 It is a waveform diagram of the frequency spectrum signal measured when the object to be detected is a pedestrian in the embodiment of the present invention.

[0021] Description of Reference Numerals

[0022] 100: Radar detection system.

[0023] 200: Methods for determining living things.

[0024] 300: Method for establishing feature database.

[0025] 10: Image analysis unit.

[0026] 20: Spectrum analysis unit.

[0027] 30: Arithmetic processing unit.

[0028] 31: Feature extraction unit.

[0029] 32: Matching determination unit.

[0030] 40: Feature database.

[0031] 41: Life form sub-database.

[0032] 42: Non-living organisms sub-database.

[0033] 50: Radar detection module.

[0034] 51: Point cloud radar.

[0035] 52: Micro Doppler radar.

[0036] 53: Continuous wave radar.

[0037] F1: feature to be measured.

[0038] F2: Control features.

[0039] F3: Control features.

[0040] P: sampling point.

[0041] S1: Signal.

[0042] S1A: First signal.

[0043] S1B: Second signal.

[0044] S2: Image signal.

[0045] S3: Spectrum signal.

[0046] S4: Image sample signal.

[0047] S5: spectrum sample signal. DETAILED DESCRIPTION

[0048] Exemplary embodiments of the present invention are described in detail below with reference to the accompanying drawings, without intending to limit the technical principles of the present invention to the specifically disclosed embodiments.

[0049] See also Figures 1 to 10 As shown, the present invention provides a radar detection system 100 for detecting a living being, a method 200 for determining a living being, and a method 300 for establishing a feature database.

[0050] The radar detection system 100 for detecting life forms, such as Figure 1As shown, in this embodiment, an image analysis unit 10, a spectrum analysis unit 20 and a calculation processing unit 30 are included, wherein the calculation processing unit 30 further includes a feature extraction unit 31 and a matching determination unit 32, wherein the matching determination unit 32 is electrically connected to a feature database 40. In this embodiment, the calculation processing unit 30 uses an artificial intelligence system of a long short-term memory model (Long Sort-Term Memory, LSTM) algorithm to perform feature extraction and comparison judgment, but is not limited thereto.

[0051] In this embodiment, a radar detection module 50 is further included. The radar detection module 50 can transmit and receive electromagnetic signals, and the wavelength of the electromagnetic signals can meet the following conditions: 0.001 mm < λ ≤ 12 mm. Preferably, it can meet the following conditions: 1 mm < λ ≤ 10 mm. More preferably, it can meet the following conditions: 3 mm < λ ≤ 5 mm. The radar detection module 50 includes a point cloud radar 51 and a micro Doppler radar 52. The point cloud radar 51 detects an object to be detected and generates a first signal S1A. The micro Doppler radar 52 detects the same object to be detected and generates a second signal S1B.

[0052] The radar detection system 100 for detecting life forms, such as Figure 3 An example of a method 200 for determining a living being is given. In this embodiment, the image analysis unit 10 receives a signal S1A and processes the signal S1A to generate a plurality of sampling points P and form an image signal S2. In addition, the spectrum analysis unit 20 receives a signal S1B and processes the signal S1B to generate a spectrum signal S3. Furthermore, the feature extraction unit 31 receives the image signal S2 and the spectrum signal S3 and performs feature extraction to obtain a feature F1 to be detected of the object to be detected, and the matching determination unit 32 compares the feature F1 to be detected to determine whether the object to be detected is a living being.

[0053] The radar detection module 50 is not limited to the two radars of the point cloud radar 51 and the micro-Doppler radar 52 described in this embodiment. In different embodiments, the radar detection module 50 can also be a single continuous wave radar 53 (such as Figure 2As shown in the figure, the electromagnetic signal that can be sent and received by the continuous wave radar 53 has a wavelength that satisfies the following conditions: 0.001 mm < λ≤ 10 mm. Preferably, it satisfies the following conditions: 1 mm < λ ≤ 10 mm. More preferably, it satisfies the following conditions: 3 mm < λ ≤ 5 mm. The continuous wave radar 53 detects an object to be detected and generates a signal S1. The signal S1 is respectively formed into the aforementioned image signal S2 and spectrum signal S3 through the image analysis unit 10 and the spectrum analysis unit 20, and can also be used by the feature extraction unit 31 for feature extraction.

[0054] The feature database 40 stores a reference feature F2 of a living being. When the matching determination unit 32 compares the feature to be detected F1, the matching determination unit 32 accesses the reference feature F2 and compares the feature to be detected F1 with the reference feature F2.

[0055] The radar detection system 100 for detecting life forms, such as Figure 4 An example of a feature database establishment method 300 is provided, in which a radar detection module 50 detects an object to be detected. In this embodiment, a point cloud radar 51 is used to detect the object to be detected to obtain an image sample signal S4, and a micro-Doppler radar 52 is used to detect the object to be detected to obtain a spectrum sample signal S5. The feature extraction unit 31 receives the image sample signal S4 and the spectrum sample signal S5 and performs feature extraction to obtain a control feature. If the control feature is determined to be a living object after comparison, it is established in the feature database 40 and classified as a control feature F2 of a living object; if the control feature is determined to be a non-living object after comparison, it is established in the feature database 40 and classified as a control feature F3 of a non-living object.

[0056] In a preferred embodiment, the feature database 40 includes a living body sub-database 41, and the living body sub-database 41 stores the control feature F2 classified as a living body, which is used as a standard for the matching judgment unit 32 to determine that the object to be detected is a living body. Further, the feature database 40 also includes a non-living body sub-database 42, and the non-living body sub-database 42 stores the control feature F3 classified as a non-living body, which is used as a standard for the matching judgment unit 32 to exclude the object to be detected as a living body. The non-living body mentioned here is mainly an object without a moving feature, such as a tree or a telephone pole.

[0057] In the method 200 for determining a living being of the present embodiment, the living being is a pedestrian, for example. When the reference feature F2 is compared with the feature to be detected F1, the image signal S2 is compared with the lower limb movement of the human body when walking on two feet of the image sample signal S4, and the spectrum signal S3 is compared with the lower limb and trunk movement of the human body when walking on two feet of the spectrum sample signal S5. If the object to be detected is another living being (such as a dog or other animal) instead of a pedestrian, or is a non-living being (such as a tree or a telephone pole), the difference between the object and the pedestrian is determined by the shape of the lower limb and trunk.

[0058] The following is a description of the radar detection system 100 for detecting life forms and the method 200 for determining life forms according to an embodiment of the present invention, which are applied to determining objects to be detected during vehicle driving.

[0059] When a vehicle equipped with the above-mentioned radar detection system 100 for detecting life forms is driving on the road, for example, in one embodiment, the radar detection module 50 detects a "tree" as an object to be detected, that is, the point cloud radar 51 generates a first signal S1A, and the micro-Doppler radar 52 generates a second signal S1B. At this time, the image analysis unit 10 receives the signal S1A and generates an image signal S2 after signal processing. The multiple sampling points P of the image signal S2 are as follows: Figure 5 As shown, it includes an obvious straight trunk part and scattered sampling points P located at the top of the straight trunk. As can be seen from the dynamic changes of the sampling points P, there is no movement of the lower limbs in the straight trunk part and it is fixed. Only the scattered sampling points P located at the top of the straight trunk move slowly. The spectrum analysis unit 20 receives the signal S1B and performs signal processing to generate a spectrum signal S3, as shown in FIG. Figure 6 The waveform diagram only shows a gentle wave. Therefore, the feature extraction unit 31 extracts the feature F1 of the object to be detected from the image signal S2 and the spectrum signal S3, and after comparing it with the control feature F2, the object to be detected with the feature F1 is not a pedestrian or animal. After comparing it with the control feature F3, it can be judged from the features of the image signal S2 and the spectrum signal S3 of the feature F1 that the object to be detected is not a living body.

[0060] In another embodiment, when the radar detection module 50 detects a dog as an object to be detected, the image analysis unit 10 receives the signal S1A and generates an image signal S2 after performing signal processing as described above. The multiple sampling points P of the image signal S2 are as follows: Figure 7As shown in FIG. 1 , the horizontal trunk portion is obviously lower in height and the sampling point P at the bottom of the horizontal trunk is similar to the lower limbs. As can be seen from the dynamic changes of the sampling point P, there is a staggered movement of four feet in the lower limbs. The spectrum analysis unit 20 receives the signal S1B and performs signal processing to generate a spectrum signal S3, as shown in FIG. Figure 8 The waveform diagrams shown, which represent the waveform of the horizontal trunk part and the waveform of the lower limbs, all have obvious irregular waves generated at the peaks and troughs. Therefore, the feature extraction unit 31 extracts the feature F1 of the object to be detected by the image signal S2 and the spectrum signal S3, and after comparing it with the control feature F2, the object to be detected with the feature F1 is a living body. Furthermore, the object to be detected can also be judged as a dog according to the features of the image signal S2 and the spectrum signal S3.

[0061] In another embodiment, when the radar detection module 50 detects a pedestrian, the image analysis unit 10 receives the signal S1A and processes the signal to generate an image signal S2. The multiple sampling points P of the image signal S2 are as follows: Fig. 9 As shown, it includes a straight trunk part and sampling points P distributed at the lower limbs and upper limbs of the straight trunk. The regular movement of the lower limbs can be clearly seen through the dynamic changes of the sampling points P; the spectrum analysis unit 20 receives the signal S1B and performs signal processing to generate a spectrum signal S3, as shown in FIG. Fig.10 The waveform diagram shown in the figure shows the waveform of the straight trunk and the waveform of the lower limbs. The two waveforms showing the straight trunk and the lower limbs respectively show regular sine wave distribution. Therefore, the feature extraction unit 31 extracts the feature F1 of the object to be detected by the image signal S2 and the spectrum signal S3. After comparing with the control feature F2, the object to be detected with the feature F1 is a living body. Furthermore, the object to be detected can also be judged as a pedestrian according to the features of the image signal S2 and the spectrum signal S3.

[0062] It is further explained here that if it is necessary to further determine whether the object to be detected is a "pedestrian", in the above-mentioned examples where the object to be detected is a tree, a dog or a pedestrian, the features of each object to be detected can still be clearly determined. Regardless of the image signal S2 or the spectrum signal S3, the feature extraction unit 31 performs feature extraction and then the matching determination unit 32 compares the result, and can accurately determine whether the object to be detected is a pedestrian or a non-pedestrian. However, in some cases, it is very likely that the image signal S2 or the spectrum signal S3 will make the tree or the dog look very similar to a pedestrian, thereby causing the possibility of misjudgment.

[0063] The above-mentioned misjudgment situation occurs when, for example, the trunk of a tree sways slightly due to wind, while the branches sway more greatly, which is similar to the situation when a person moves slowly: the trunk sways slightly, while the limbs sway more greatly, so the performance of the spectrum signal S3 will be similar, causing difficulty in judgment; but if in the performance of the image signal S2, because the image of the tree has an obvious straight trunk part and scattered sampling points located at the top of the straight trunk, and there is only one continuous point in the lower limb part, which is different from the lower limbs of a person having obvious two legs, it can be clearly determined through the image signal S2 that the object to be detected is not a pedestrian.

[0064] In terms of the dog's judgment mechanism, when the dog slowly walks toward or away from the vehicle, the image signal S2 shows that the lower limbs are similar to the situation of pedestrians walking with two feet. Only two feet can be seen, or even because the lower limbs are too close, only a continuous dot pattern can be presented, and the four feet cannot be clearly seen. Moreover, the height is similar to that of a human child. At this time, it is impossible to clearly judge whether it is a pedestrian through the image signal S2; but if the spectrum signal S3 shows that the time period and shape of the swing of the four limbs are different from those of pedestrians, the spectrum signal S3 can be clearly determined to be a non-pedestrian. Therefore, the present invention receives the image signal S2 and the spectrum signal S3 by the operation processing unit 30, extracts features through the feature extraction unit 31, and then compares them through the matching judgment unit 32, which can indeed achieve the effect of complementary judgment and greatly reduce the possibility of misjudgment.

[0065] It is not difficult to find out from the above description that the main features of the present invention are:

[0066] 1. By performing feature extraction on the image signal S2 and the spectrum signal S3 by the processing unit 30, it is possible to quickly and accurately determine whether it is a living being. Compared with the existing method of using images to determine whether it is a pedestrian, it is not necessary to perform complex algorithms to confirm the features to be tested, and it also avoids the problem of image misjudgment caused by light interference when using a camera to read the image.

[0067] 2. Through the setting of the feature database 40, when the matching judgment unit 32 compares the feature to be detected, it can use the control feature F2 classified as a living body as the comparison standard, so as to provide a basis for judging whether the object to be detected is a living body. In addition, the feature database 40 can continuously establish the control feature F2 classified as a living body in the living body sub-database 41, or establish the control feature F3 classified as a non-living body in the non-living body sub-database 42, so as to achieve artificial intelligence deep learning, so that when the matching judgment unit 32 compares the feature F1 to be detected, there will be sufficient comparison samples as a reference, thereby improving the judgment accuracy.

[0068] 3. The signal S1 generated by a single continuous wave radar 53 can be used for the image analysis unit 10 and the spectrum analysis unit 20 to form an image signal S2 and a spectrum signal S3 respectively, so as to be used for feature extraction by the feature extraction unit 31, thereby achieving the effect of reducing the volume and lowering the radar setting cost.

[0069] The above embodiments are only used to illustrate the present invention, and are not used to limit the protection scope of the present invention. Any modifications or changes that do not violate the present invention are within the scope of the present invention.

Claims

1. A radar detection system for a living being, characterized in that: Include: An image analysis unit receives a signal measured by a radar and performs signal processing on the signal to generate a plurality of sampling points to form an image signal; a spectrum analysis unit, receiving the signal and performing signal processing on the signal to generate a spectrum signal; and a processing unit electrically connected to the image analysis unit and the spectrum analysis unit, the processing unit comprising a feature extraction unit and a matching determination unit, the feature extraction unit receiving the image signal and the spectrum signal to extract features to obtain a feature to be detected of an object to be detected, and the matching determination unit comparing the feature to be detected; The matching determination unit is electrically connected to a feature database, and the feature database stores a reference feature of a living body. When the matching determination unit compares the feature to be tested, the matching determination unit accesses the reference feature and compares the feature to be tested with the reference feature. The control feature is obtained by extracting features from an image sample signal and a spectrum sample signal obtained by radar detection of a living body; Wherein, the living body is a human body, and when the control feature is compared with the feature to be measured, the image signal is compared with the movement of the lower limbs of the human body when walking on two feet of the image sample signal, and the spectrum signal is compared with the movement of the lower limbs and torso of the human body when walking on two feet of the spectrum sample signal.

2. The radar detection system for a living being as claimed in claim 1, characterized in that: The computing processing unit uses an artificial intelligence system of a long short-term memory model algorithm to perform feature extraction and comparison judgment.

3. The radar detection system for a living being as claimed in claim 1, characterized in that: It further includes a radar detection module to detect the object to be detected and generate the signal.

4. The radar detection system for a living being as claimed in claim 3, characterized in that: The radar detection module includes a point cloud radar and a micro Doppler radar. The signal generated by the point cloud radar is passed through the image analysis unit to form the image signal, and the signal generated by the micro Doppler radar is passed through the spectrum analysis unit to generate the spectrum signal.

5. The radar detection system for a living being as claimed in claim 3, characterized in that: The radar detection module has a continuous wave radar, and the signal generated by the continuous wave radar is respectively formed into the image signal and the spectrum signal through the image analysis unit and the spectrum analysis unit.

6. A method for determining a living being in a living being radar detection system, characterized in that: Include: An image analysis unit receives a signal detected by a radar, and performs signal processing on the signal to generate a plurality of sampling points to form an image signal; A spectrum analysis unit is used to receive the signal and perform signal processing on the signal to generate a spectrum signal; Using a feature extraction unit to perform feature extraction on the image signal and the spectrum signal to obtain a feature to be detected of an object to be detected; and Using a matching determination unit to compare the feature to be detected to determine whether the object to be detected is a living being; When the matching determination unit compares the feature to be detected, the matching determination unit accesses a control feature of a living body and compares the feature to be detected with the control feature; The control feature is obtained by extracting features from an image sample signal and a spectrum sample signal obtained by radar detection of a living body; Wherein, the living body is a human body, and when the control feature is compared with the feature to be measured, the image signal is compared with the movement of the lower limbs of the human body when walking on two feet of the image sample signal, and the spectrum signal is compared with the movement of the lower limbs and torso of the human body when walking on two feet of the spectrum sample signal.

7. The method for determining a living being in a living being radar detection system according to claim 6, characterized in that: The comparison feature is established in a feature database and classified as a comparison feature of a living body, so as to serve as a standard for the matching determination unit to determine that the object to be detected is a living body.

8. The method for determining a living being in a living being radar detection system according to claim 7, characterized in that: The feature database includes a living body sub-database and a non-living body sub-database. The control features of the living body are stored in the living body sub-database; the non-living body sub-database stores the control features classified as non-living bodies to serve as the standard for the matching judgment unit to exclude the object to be detected as a living body.

9. A method for establishing a feature database of a life form radar detection system according to claim 1, wherein a radar signal is collected from a life form, and the method is characterized in that: The feature extraction unit is electrically connected to the image analysis unit and the spectrum analysis unit. The feature extraction unit extracts features from the image sample signal and the spectrum sample signal to obtain a comparison feature, which is then established in a feature database and classified as a comparison feature of a living being.

10. The method for establishing a feature database of a life form radar detection system as claimed in claim 9, characterized in that: The feature database includes a living body sub-database and a non-living body sub-database. The control features of the living body are stored in the living body sub-database. The control features that are judged to be non-living bodies after comparison are established in the non-living body sub-database and classified as control features of non-living bodies.

Images