CN109960278B - LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle - Google Patents
LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle Download PDFInfo
- Publication number
- CN109960278B CN109960278B CN201910281845.5A CN201910281845A CN109960278B CN 109960278 B CN109960278 B CN 109960278B CN 201910281845 A CN201910281845 A CN 201910281845A CN 109960278 B CN109960278 B CN 109960278B
- Authority
- CN
- China
- Prior art keywords
- lgmd
- membrane potential
- layer
- neuron
- obstacle avoidance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 239000011664 nicotinic acid Substances 0.000 title claims abstract description 19
- 201000009342 Limb-girdle muscular dystrophy Diseases 0.000 title claims description 47
- RZVHIXYEVGDQDX-UHFFFAOYSA-N 9,10-anthraquinone Chemical compound C1=CC=C2C(=O)C3=CC=CC=C3C(=O)C2=C1 RZVHIXYEVGDQDX-UHFFFAOYSA-N 0.000 claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 claims abstract description 24
- 238000004891 communication Methods 0.000 claims abstract description 21
- 230000003287 optical effect Effects 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 14
- 210000002569 neuron Anatomy 0.000 claims description 157
- 239000012528 membrane Substances 0.000 claims description 151
- 230000005764 inhibitory process Effects 0.000 claims description 16
- 230000001537 neural effect Effects 0.000 claims description 16
- 230000001629 suppression Effects 0.000 claims description 13
- 230000002401 inhibitory effect Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000000007 visual effect Effects 0.000 claims description 9
- 230000002708 enhancing effect Effects 0.000 claims description 6
- 230000002964 excitative effect Effects 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 230000004888 barrier function Effects 0.000 claims description 5
- 230000005284 excitation Effects 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- 150000001875 compounds Chemical class 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 6
- 239000003550 marker Substances 0.000 abstract 1
- 238000011160 research Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 241000238631 Hexapoda Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000000638 stimulation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/048—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators using a predictor
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- G05D1/0088—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
Abstract
The invention provides an LGMD (laser marker detection) -based bionic obstacle avoidance control system of an unmanned aerial vehicle, which comprises a flight control subsystem, an optical flow sensor, a driving motor, an embedded LGMD detector, a camera, a wireless communication module and a ground station PC (personal computer); the optical flow sensor and the embedded LGMD detector are electrically connected with the flight control subsystem; the wireless communication module and the driving motor are electrically connected with the output end of the flight control subsystem; the input end of the embedded LGMD detector is in signal connection with the output end of the camera; the wireless communication module is in wireless communication connection with the ground station PC. The invention further provides an LGMD-based bionic obstacle avoidance control method for the unmanned aerial vehicle, which is characterized in that an LGMD neural network is built, and a field image is segmented, so that space direction selection and scene prediction are realized in the flight process of the unmanned aerial vehicle, and real-time and efficient obstacle avoidance flight of the unmanned aerial vehicle in an unknown environment is realized.
Description
Technical Field
The invention relates to the technical field of unmanned aerial vehicle aircrafts, in particular to an LGMD (light-gauge mechanical system) -based unmanned aerial vehicle bionic obstacle avoidance control system and an LGMD-based unmanned aerial vehicle bionic obstacle avoidance control method.
Background
Unmanned aerial vehicles have wide application prospects in a plurality of scenes such as geographic measurement, agricultural aviation, danger detection and the like, and safety is always the focus of attention of people, particularly in complex environments. Traditional unmanned aerial vehicles use GPS and optical flow for path planning and detection of collision in combination with detection methods of sensors such as ultrasonic waves, infrared rays and laser, however, such methods depend greatly on the complexity of barrier materials, textures and backgrounds and can only be used in simple and specific environments. In recent years, the obstacle avoidance method based on biological vision is a hot spot for studying by scholars at home and abroad due to high efficiency and flexibility. At present, foreign scholars have performed a great deal of biological experiments on insect visual navigation and obstacle avoidance mechanisms, and deep research summary is performed to form a great deal of borrowable research results. Anatomical experiments on a locust visual system prove that a Lobular Giant Movement Detector (LGMD) is a main neuron for finishing a collision early warning function, and Rind F.C. and Bramwell D.I. and the like provide a classic 4-layer LGMD input neural network structure, adopt an excitation inhibition mode to detect the approach of an object, and start the research of collision early warning by means of the LGMD network. Yue S, et al, introduce new neurons to enhance the LGMD collision early warning capability, prove the validity and robustness of the LGMD network to the collision early warning, and transplant the LGMD network to a micro-robot to study the collision early warning capability under the population condition. However, at present, the application research of the LGMD on the unmanned aerial vehicle is less, random selection is still adopted in the aspect of selecting the obstacle avoidance direction, and prediction is lacked in unknown scenes.
Disclosure of Invention
The invention provides an LGMD-based unmanned aerial vehicle bionic obstacle avoidance control system for overcoming the technical defects that the traditional unmanned aerial vehicle collision detection method depends on the complexity of barrier materials, textures and backgrounds and can only be used in simple and specific environments, and improving the flexibility and efficiency of unmanned aerial vehicle obstacle avoidance.
And the system also provides an LGMD-based bionic obstacle avoidance control method for the unmanned aerial vehicle.
In order to solve the technical problems, the technical scheme of the invention is as follows:
an LGMD-based bionic obstacle avoidance control system of an unmanned aerial vehicle comprises a flight control subsystem, an optical flow sensor, a driving motor, an embedded LGMD detector, a camera, a wireless communication module and a ground station PC; wherein:
the optical flow sensor and the embedded LGMD detector are electrically connected with the flight control subsystem for information interaction;
the wireless communication module and the driving motor are electrically connected with the output end of the flight control subsystem;
the input end of the embedded LGMD detector is in signal connection with the output end of the camera;
the wireless communication module is in wireless communication connection with the ground station PC.
The embedded LGMD detector is provided with an LGMD neural network, video information collected by the camera is calculated through the LGMD neural network, an obstacle avoidance control instruction is obtained through the LGMD neural network and output to the flight control subsystem, and obstacle avoidance control of the unmanned aerial vehicle is achieved.
Wherein the LGMD neural network comprises P layer neurons, E layer neurons, I layer neurons, S layer neurons, G layer neurons, LGMD neurons, and feed forward inhibitory FFI neurons; wherein:
the P-layer neuron acquires the field image information of an input video and responds to frame difference to obtain a P-layer neuron membrane potential;
the P layer neuronal membrane potential acts directly as the excitatory membrane potential of the E layer neurons;
the I layer neuron receives the output of a frame on the P layer neuron and performs local inhibition to obtain an inhibition membrane potential;
the neuron of the S layer gathers the output in the corresponding position field of the neuron of the I layer, and the neuron of the E layer is inhibited through inhibiting membrane potential to obtain membrane potential of the neuron of the S layer;
the G-layer neuron is used for enhancing and extracting collision objects under a complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential;
the LGMD neuron carries out diagonal segmentation on the image information of the visual field to obtain 4 pieces of azimuth information; calculating according to the G-layer neuron membrane potential and the 4 azimuth information to obtain the membrane potential of 4 azimuth C-LGMD, and adding the membrane potentials of the 4 azimuth C-LGMD to obtain the LGMD neuron membrane potential;
the feed-forward inhibition FFI neuron directly acquires field image information from the P layer neuron and inhibits the LGMD neuron membrane potential, so that an obstacle avoidance control instruction is obtained and output to the flight control subsystem.
The system further comprises an inertial sensor, wherein the inertial sensor is integrated on the flight control subsystem and is electrically connected with the flight control subsystem.
The system further comprises a laser sensor, and the laser sensor is electrically connected with the input end of the optical flow sensor.
In the scheme, the core of the flight control subsystem is STM32F 407V; the inertial sensor main body is an MPU6050 chip; the main body of the optical Flow sensor is a Pix4Flow chip; the main body of the wireless communication module is nRF24L 01; the inertial sensor is used for recording attitude information of the unmanned aerial vehicle, the optical flow sensor is used as a horizontal plane position and speed feedback device, and the camera is used for collecting video information in real time; the flight control subsystem calculates PWM values corresponding to the driving motors according to instructions output by the embedded LGMD detector, and finally outputs the PWM values to four driving motors of the unmanned aerial vehicle to realize obstacle avoidance control; and the wireless communication module returns real-time data to the ground station PC.
An LGMD-based bionic obstacle avoidance control method for an unmanned aerial vehicle comprises the following steps:
s1: performing real-time video acquisition through a camera to obtain an input video;
s2: the embedded LGMD detector acquires the view field image information of an input video, obtains an obstacle avoidance control instruction through LGMD neural network calculation, and outputs the obstacle avoidance control instruction to the flight control subsystem to realize the obstacle avoidance control of the unmanned aerial vehicle.
The specific process of calculating the obstacle avoidance control instruction by the LGMD neural network comprises the following steps:
s21: the P-layer neuron acquires the visual field image information of the input video, responds to the frame difference and obtains the membrane potential P of the P-layer neuronf(x, y), specifically:
Pf(x,y)=Lf(x,y)-Lf-1(x,y); 1)
wherein: f represents the f-th frame in the video sequence, (x, y) is the position of the pixel point in the network layer, Lf(x, y) are pixel values of the input field-of-view image;
s22: the output of the P layer neuron is used as the input of the E layer neuron and the I layer neuron, the E layer neuron directly receives the output of the P layer neuron, and the I layer neuron receives the output of a frame on the P layer neuron and performs local inhibition, which is specifically represented as:
Ef(x,y)=Pf(x,y); 2)
If(x,y)=∑i∑jPf-1(x+i,y+j)wi(i,j)(if i=j,j≠0); 3)
wherein: ef(x, y) is the excitatory membrane potential, i.e.the E-layer neuronal membrane potential; i isf(x, y) is the inhibitory membrane potential, i.e., the layer I neuronal membrane potential; w is ai(i, j) is the local suppression weight; i and j are not 0 at the same time;
s23: the method comprises the following steps that (1) the neurons in the S layer converge output in the corresponding position field of the neurons in the I layer, and the neurons in the E layer are inhibited through inhibiting membrane potential to obtain the membrane potential of the neurons in the S layer, and specifically the membrane potential of the neurons in the S layer is as follows:
Sf(x,y)=Ef(x,y)-If(x,y)WI; 5)
wherein: sf(x, y) is the membrane potential of S-layer neurons; wIIs a suppression weight matrix;
s24: the G-layer neuron is used for enhancing and extracting collision objects under the complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential, and the method specifically comprises the following steps:
wherein: gf(x, y) is G-layer neuronal membrane potential; [ w ]e]Is a convolution kernel; r is the convolution kernel radius, and r is 1;
setting a threshold TdeTo filter weak excitation points, specifically:
wherein:is a filtered G-layer neuronal membrane potential, CdeHas a coefficient of weakness of [0, 1%];TdeIs a filtering threshold;
s25: when the field image information reaches the LGMD neuron, the field image information is diagonally segmented to obtain coordinates on a y axis of two diagonal lines, which are specifically expressed as:
wherein: diag1, Diag2 are y coordinates of x corresponding to two diagonal lines, respectively; w is the width of the image; h is the height of the image; thereby obtaining the membrane potential of the 4-direction C-LGMD, which is specifically as follows:
wherein: u shapeLGMD,DLGMD,LLGMD,RLGMDMembrane potentials of 4 directions of upper, lower, left and right of the image respectively; adding the membrane potentials of the 4 orientations C-LGMD to obtain the LGMD neuron membrane potential KfThe method specifically comprises the following steps:
will KfNormalized and mapped to [0,255 ]]The range of (a) is specifically:
wherein n iscellThe total number of pixel points of the image;
obtaining the membrane potential after mapping in 4 directions according to the proportion of the membrane potential of the C-LGMD in the whole image in the 4 directions, which specifically comprises the following steps:
wherein the content of the first and second substances,respectively mapping the membrane potentials of the upper, lower, left and right 4 azimuths of the image; when k isfExceeds its threshold value TsThen an LGMD peak pulse is generatedThe method specifically comprises the following steps:
if n is continuoustsPulse is not less than nspThen, it is determined that a collision is about to occur, which is specifically expressed as:
s26: the feedforward inhibition FFI neuron directly acquires field-of-view image information from the P layer neuron, which is specifically represented as:
wherein, FfInhibiting the membrane potential of an FFI neuron for feedforward; t isFFIIs a preset threshold value; when F is presentfExceeds a threshold value TFFILGMD neuronThe membrane potential is immediately suppressed;
s27: selecting the direction: FFI membrane potential F is inhibited by current feedfGreater than a threshold value TFFIAll obstacle avoidance instructions are invalid; when C is presentfinalTurn and feed forward suppression of FFI membrane potential FfLess than threshold TFFIThen, the obstacle is judged to appear, and 4 directions are comparedThe magnitude of the membrane potential takes the direction with the minimum membrane potential among the 4 directions as the safest obstacle avoidance direction;
s28: predicting a flight scene: when C is presentfinalFALSE and feed forward suppression of FFI Membrane potential FfLess than threshold TFFIWhen the unmanned aerial vehicle flies normally, predicting an unknown flying scene of the unmanned aerial vehicle by acquiring membrane potentials of 4 orientations of the FFI;
s29: and forming an obstacle avoidance control command by using the signals obtained in the steps S25, S27 and S28 as obstacle avoidance control signals.
The flight scene prediction process in step S28 specifically includes:
setting a scene prediction threshold TFFIIFFI average membrane potential when N frames are aheadLess than threshold TFFIIIn time, the unmanned aerial vehicle flies normally; FFI average membrane potential when N frames are aheadGreater than a threshold value TFFIIIs less than a threshold value TFFIBy comparisonFinding out the direction with the minimum membrane potential so as to predict the direction with the minimum future flight direction barrier in the front scene;
wherein, the FFI average membrane potential is calculated by N frames aheadThe concrete expression is as follows:
in the formula (I), the compound is shown in the specification,represents the average value of the membrane potential of N frames before 4 orientations of FFI, including the upper, lower, left and right orientations.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
according to the system and the method for the bionic obstacle avoidance control system of the unmanned aerial vehicle based on the LGMD, provided by the invention, the LGMD neural network is built, and the field image is segmented, so that the space direction selection and the scene prediction are realized in the flight process of the unmanned aerial vehicle, and the real-time and efficient obstacle avoidance flight of the unmanned aerial vehicle in an unknown environment is realized.
Drawings
FIG. 1 is a schematic diagram of the structural connections of the system of the present invention;
FIG. 2 is a schematic diagram of an LGMD neural network;
FIG. 3 is a schematic flow diagram of the process of the present invention;
FIG. 4 is a view of field image segmentation;
wherein: 1. a flight control subsystem; 2. an optical flow sensor; 3. a drive motor; 4. an embedded LGMD detector; 5. a camera; 6. a wireless communication module; 7. a ground station PC; 8. a laser sensor.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
As shown in fig. 1, an LGMD-based unmanned aerial vehicle bionic obstacle avoidance control system includes a flight control subsystem 1, an optical flow sensor 2, a driving motor 3, an embedded LGMD detector 4, a camera 5, a wireless communication module 6 and a ground station PC 7; wherein:
the optical flow sensor 2 and the embedded LGMD detector 4 are electrically connected with the flight control subsystem 1 for information interaction;
the wireless communication module 6 and the driving motor 3 are electrically connected with the output end of the flight control subsystem 1;
the input end of the embedded LGMD detector 4 is in signal connection with the output end of the camera 5;
the wireless communication module 6 is in wireless communication connection with a ground station PC 7.
More specifically, the embedded LGMD detector 4 is provided with an LGMD neural network, video information acquired by the camera 5 is calculated through the LGMD neural network, and an obstacle avoidance control instruction is obtained by the LGMD neural network and output to the flight control subsystem 1, so that obstacle avoidance control of the unmanned aerial vehicle is realized.
More specifically, as shown in fig. 2, the LGMD neural network includes P-layer neurons, E-layer neurons, I-layer neurons, S-layer neurons, G-layer neurons, LGMD neurons, and feed-forward inhibitory FFI neurons; wherein:
the P-layer neuron acquires the field image information of an input video and responds to frame difference to obtain a P-layer neuron membrane potential;
the P layer neuronal membrane potential acts directly as the excitatory membrane potential of the E layer neurons;
the I layer neuron receives the output of a frame on the P layer neuron to obtain an inhibition membrane potential;
the neuron of the S layer gathers the output in the corresponding position field of the neuron of the I layer, and the neuron of the E layer is inhibited through inhibiting membrane potential to obtain membrane potential of the neuron of the S layer;
the G-layer neuron is used for enhancing and extracting collision objects under a complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential;
the LGMD neuron carries out diagonal segmentation on the image information of the visual field to obtain 4 pieces of azimuth information; calculating according to the G-layer neuron membrane potential and the 4 azimuth information to obtain the membrane potential of 4 azimuth C-LGMD, and adding the membrane potentials of the 4 azimuth C-LGMD to obtain the LGMD neuron membrane potential;
the feed-forward inhibition FFI neuron directly acquires field image information from the P layer neuron, and inhibits the LGMD neuron membrane potential, so that an obstacle avoidance control instruction is obtained and output to the flight control subsystem 1.
More specifically, the system further comprises an inertial sensor integrated on the flight control subsystem and electrically connected with the flight control subsystem.
More specifically, the system still includes laser sensor 8, laser sensor 8 with optical flow sensor 2 input electric connection, with optical flow sensor 2 cooperation is used, mainly used unmanned aerial vehicle's height finding, height-fixing.
In a specific implementation process, the core of the flight control subsystem 1 is STM32F 407V; the inertial sensor main body is an MPU6050 chip; the main body of the optical Flow sensor 2 is a Pix4Flow chip; the main body of the wireless communication module 6 is nRF24L 01; the inertial sensor is used for recording attitude information of the unmanned aerial vehicle, the optical flow sensor 2 is used as a horizontal plane position and speed feedback device, and the camera 5 is used for collecting video information in real time; the flight control subsystem 1 calculates the corresponding PWM value of the driving motor 3 according to the instruction output by the embedded LGMD detector 4, and finally outputs the PWM value to the four driving motors 3 of the unmanned aerial vehicle to realize obstacle avoidance control; the wireless communication module 6 returns real-time data to the ground station PC 7.
Example 2
More specifically, on the basis of embodiment 1, an LGMD-based unmanned aerial vehicle bionic obstacle avoidance control method is provided, which includes the following steps:
s1: performing real-time video acquisition through a camera 5 to obtain an input video;
s2: the embedded LGMD detector 4 acquires the view field image information of an input video, obtains an obstacle avoidance control instruction through LGMD neural network calculation, and outputs the obstacle avoidance control instruction to the flight control subsystem 1 to realize the obstacle avoidance control of the unmanned aerial vehicle.
More specifically, as shown in fig. 3, the specific process of calculating the obstacle avoidance control instruction by the LGMD neural network includes:
s21: the P-layer neuron acquires the visual field image information of the input video, responds to the frame difference and obtains the membrane potential P of the P-layer neuronf(x, y), specifically:
Pf(x,y)=Lf(x,y)-Lf-1(x,y); 1)
wherein: f represents the f-th frame in the video sequence, (x, y) is the position of the pixel point in the network layer, Lf(x, y) are pixel values of the input field-of-view image;
s22: the output of the P layer neuron is used as the input of the E layer neuron and the I layer neuron, the E layer neuron directly receives the output of the P layer neuron, and the I layer neuron receives the output of a frame on the P layer neuron and performs local inhibition, which is specifically represented as:
Ef(x,y)=Pf(x,y); 2)
If(x,y)=∑i∑jPf-1(x+i,y+j)wi(i,j)(if i=j,j≠0); 3)
wherein: ef(x, y) is the excitatory membrane potential, i.e.the E-layer neuronal membrane potential; i isf(x, y) is the inhibitory membrane potential, i.e., the layer I neuronal membrane potential; w is ai(i, j) is the local suppression weight; i and j are not 0 at the same time, and the local suppression weight matrix of the embodiment is only one form.
S23: the method comprises the following steps that (1) the neurons in the S layer converge output in the corresponding position field of the neurons in the I layer, and the neurons in the E layer are inhibited through inhibiting membrane potential to obtain the membrane potential of the neurons in the S layer, and specifically the membrane potential of the neurons in the S layer is as follows:
Sf(x,y)=Ef(x,y)-If(x,y)WI; 5)
wherein: sf(x, y) is the membrane potential of S-layer neurons; wIIs a suppression weight matrix;
s24: the G-layer neuron is used for enhancing and extracting collision objects under the complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential, and the method specifically comprises the following steps:
wherein: gf(x, y) is G-layer neuronal membrane potential; [ w ]e]The convolution kernel and the radius thereof in the embodiment are only one form; r is the convolution kernel radius, and r is 1;
setting a threshold TdeTo filter weak excitation points, specifically:
wherein:is a filtered G-layer neuronal membrane potential, CdeHas a coefficient of weakness of [0, 1%];TdeIs a filtering threshold;
s25: when the field image information reaches the LGMD neurons, the field image information is diagonally segmented, as shown in fig. 4, and the coordinates on the y-axis of the two diagonal lines are obtained as specifically expressed as:
wherein: diag1, Diag2 are y coordinates of x corresponding to two diagonal lines, respectively; w is the width of the image; h is the height of the image; thereby obtaining the membrane potential of the 4-direction C-LGMD, which is specifically as follows:
wherein: u shapeLGMD,DLGMD,LLGMD,RLGMDMembrane potentials of 4 directions of upper, lower, left and right of the image respectively; adding the membrane potentials of the 4 orientations C-LGMD to obtain the LGMD neuron membrane potential KfThe method specifically comprises the following steps:
will KfNormalized and mapped to [0,255 ]]The range of (a) is specifically:
wherein n iscellThe total number of pixel points of the image;
obtaining the membrane potential after mapping in 4 directions according to the proportion of the membrane potential of the C-LGMD in the whole image in the 4 directions, which specifically comprises the following steps:
wherein the content of the first and second substances,respectively mapping the membrane potentials of the upper, lower, left and right 4 azimuths of the image; when k isfExceeds its threshold value TsThen an LGMD peak pulse is generatedThe method specifically comprises the following steps:
if n is continuoustsPulse is not less than nspThen, it is determined that a collision is about to occur, which is specifically expressed as:
s26: the feedforward inhibition FFI neuron directly acquires field-of-view image information from the P layer neuron, which is specifically represented as:
wherein, FfInhibiting the membrane potential of an FFI neuron for feedforward; t isFFIIs a preset threshold value; when F is presentfExceeds a threshold value TFFIThe LGMD neuron membrane potential is immediately suppressed;
s27: selecting the direction: FFI membrane potential F is inhibited by current feedfGreater than a threshold value TFFIAll obstacle avoidance instructions are invalid; when C is presentfinalTurn and feed forward suppression of FFI membrane potential FfLess than threshold TFFIThen, the obstacle is judged to appear, and 4 directions are comparedThe magnitude of the membrane potential takes the direction with the minimum membrane potential among the 4 directions as the safest obstacle avoidance direction;
in the specific implementation process, the direction with the minimum film potential in 4 directions is taken as the safest obstacle avoidance direction, and the output function constructed by the C-LGMD film potential, the change speed and the acceleration in the obstacle avoidance direction is taken asNumber UfAnd giving out obstacle avoidance speed, collecting the calculation result of the LGMD neural network model by the motion decision neuron, outputting the calculation result as an embedded LGMD detector, and giving out a corresponding obstacle avoidance signal.
In the implementation process, when an obstacle avoidance signal is generated, and an obstacle avoidance signal is detectedIf the membrane potential is minimum, the direction is the optimal obstacle avoidance direction, and an output function U constructed by combining the membrane potential, the change speed and the acceleration is combinedfAnd giving the obstacle avoidance speed. In special cases, when the obstacle is just in the middle of the image, namely the 4 azimuthal membrane potentials are equal, the obstacle preferentially flies leftwards; if 2 or 3 membrane potentials in the same direction are present, the obstacle avoidance is preferentially carried out in the direction opposite to the direction of the strongest membrane potential.
More specifically, the output function is represented as:
wherein: k is a radical of1,k2,k3Is a proportionality coefficient;
s28: predicting a flight scene: when C is presentfinalFALSE and feed forward suppression of FFI Membrane potential FfLess than threshold TFFIWhen the unmanned aerial vehicle flies normally, predicting an unknown flying scene of the unmanned aerial vehicle by collecting 4 orientation membrane potentials of the FFI;
s29: and (5) taking the signals obtained in the steps S25, S27 and S28 as obstacle avoidance control signals to form obstacle avoidance control commands, and outputting the commands to the flight control subsystem.
More specifically, the flight scenario prediction process in step S28 specifically includes:
setting a scene prediction threshold TFFIIFFI average membrane potential when N frames are aheadLess than threshold TFFIIIn time, the unmanned aerial vehicle flies normally; FFI average membrane potential when N frames are aheadGreater than a threshold value TFFIIIs less than a threshold value TFFIBy comparisonFinding out the direction with the minimum membrane potential so as to predict the direction with the minimum future flight direction barrier in the front scene;
wherein, the FFI average membrane potential is calculated by N frames aheadThe concrete expression is as follows:
in the formula (I), the compound is shown in the specification,represents FFI upper, lower, left and rightAverage of the membrane potentials of the N frames before 4 azimuths.
In the specific implementation process, the LGMD neural network model [1] consists of 5 layers of neuron cells, namely P layer neurons, E layer neurons, I layer neurons, S layer neurons, G layer neurons, LGMD neurons and feed-forward inhibition FFI neurons, is very sensitive to fast approaching objects, and is essentially characterized in that excitation in the whole network is rapidly increased due to edge stimulation of continuous expansion of the objects when the objects approach, so that the membrane potential of the LGMD neurons is rapidly increased, and the network has the function of early warning of collision due to the change; by dividing the field-of-view image into 4 orientations: the upper, lower, left and right sides form 4-direction competitive LGMD neurons, namely C-LGMD, and the optimal obstacle avoidance direction is obtained by comparing the obstacle avoidance responses of the LGMD in different directions, so that the unmanned aerial vehicle can avoid the obstacle more efficiently and flexibly.
In the implementation process, the invention records the image information of 4 orientations of the C-LGMD in real time Further, the first derivative v is obtained from the image information of each positionU、vD、vL、vRAnd the second derivative αU、αD、αL、αRRespectively representing the velocity and acceleration of the image signal changes at each orientation. Meanwhile, accumulating the FFI membrane potential of the near N frames forward and obtaining the average valueAnd 4 values of orientation
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
[1]Yue S,Rind F C.Collision detection in complex dynamic scenes using an LGMD-based visual neural network with feature enhancement[J].IEEE Transactions on Neural Networks,2006,17(3):705-716。
Claims (6)
1. The utility model provides a bionical obstacle control system that keeps away of unmanned aerial vehicle based on LGMD which characterized in that: the system comprises a flight control subsystem (1), an optical flow sensor (2), a driving motor (3), an embedded LGMD detector (4), a camera (5), a wireless communication module (6) and a ground station PC (7); wherein:
the optical flow sensor (2) and the embedded LGMD detector (4) are electrically connected with the flight control subsystem (1) for information interaction;
the wireless communication module (6) and the driving motor (3) are electrically connected with the output end of the flight control subsystem (1);
the input end of the embedded LGMD detector (4) is in signal connection with the output end of the camera (5);
the wireless communication module (6) is in wireless communication connection with a ground station PC (7);
an LGMD neural network is arranged on the embedded LGMD detector (4), video information acquired by the camera (5) is calculated through the LGMD neural network, an obstacle avoidance control instruction is obtained through the LGMD neural network and output to the flight control subsystem (1), and obstacle avoidance control of the unmanned aerial vehicle is achieved;
the LGMD neural network comprises P layer neurons, E layer neurons, I layer neurons, S layer neurons, G layer neurons, LGMD neurons and feed forward inhibitory FFI neurons; wherein:
the P-layer neuron acquires the field image information of an input video and responds to frame difference to obtain a P-layer neuron membrane potential;
the P layer neuronal membrane potential acts directly as the excitatory membrane potential of the E layer neurons;
the I layer neuron receives the output of a frame on the P layer neuron and performs local inhibition to obtain an inhibition membrane potential;
the neuron of the S layer gathers the output in the corresponding position field of the neuron of the I layer, and the neuron of the E layer is inhibited through inhibiting membrane potential to obtain membrane potential of the neuron of the S layer;
the G-layer neuron is used for enhancing and extracting collision objects under a complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential;
the LGMD neuron carries out diagonal segmentation on the image information of the visual field to obtain 4 pieces of azimuth information; calculating according to the G-layer neuron membrane potential and the 4 azimuth information to obtain the membrane potential of 4 azimuth C-LGMD, and adding the membrane potentials of the 4 azimuth C-LGMD to obtain the LGMD neuron membrane potential;
the feed-forward inhibition FFI neuron directly acquires field image information from the P layer neuron, and inhibits the LGMD neuron membrane potential, so that an obstacle avoidance control instruction is obtained and output to the flight control subsystem (1);
when the field image information reaches the LGMD neuron, the field image information is diagonally segmented to obtain coordinates on a y axis of two diagonal lines, which are specifically expressed as:
wherein: diag1, Diag2 are y coordinates of x corresponding to two diagonal lines, respectively; w is the width of the image; h is the height of the image; thereby obtaining the membrane potential of the 4-direction C-LGMD, which is specifically as follows:
wherein:is a filtered G-layer neuronal membrane potential, ULGMD,DLGMD,LLGMD,RLGMDThe membrane potentials are respectively at the upper, lower, left and right 4 directions of the image, and the direction with the minimum membrane potential in the 4 directions is taken as the safest obstacle avoidance direction.
2. The LGMD-based bionic unmanned aerial vehicle obstacle avoidance control system is characterized in that: the flight control system is characterized by further comprising an inertial sensor, wherein the inertial sensor is integrated on the flight control subsystem (1) and is electrically connected with the flight control subsystem (1).
3. The LGMD-based bionic unmanned aerial vehicle obstacle avoidance control system is characterized in that: the device is characterized by further comprising a laser sensor (8), wherein the laser sensor (8) is electrically connected with the input end of the optical flow sensor (2).
4. An LGMD-based bionic unmanned aerial vehicle obstacle avoidance control method applied to the LGMD-based bionic unmanned aerial vehicle obstacle avoidance control system of claim 3, and characterized in that: the method comprises the following steps:
s1: real-time video acquisition is carried out through a camera (5) to obtain an input video;
s2: the embedded LGMD detector (4) acquires the view field image information of an input video, obtains an obstacle avoidance control command through LGMD neural network calculation, and outputs the obstacle avoidance control command to the flight control subsystem (1) to realize the obstacle avoidance control of the unmanned aerial vehicle.
5. The LGMD-based bionic unmanned aerial vehicle obstacle avoidance control method according to claim 4, wherein the method comprises the following steps: the specific process of calculating the obstacle avoidance control instruction by the LGMD neural network comprises the following steps:
s21: the P-layer neuron acquires the visual field image information of the input video, responds to the frame difference and obtains the membrane potential P of the P-layer neuronf(x, y), specifically:
Pf(x,y)=Lf(x,y)-Lf-1(x,y); 1)
wherein: f represents the f-th frame in the video sequence, (x, y) is the position of the pixel point in the network layer, Lf(x, y) are pixel values of the input field-of-view image;
s22: the output of the P layer neuron is used as the input of the E layer neuron and the I layer neuron, the E layer neuron directly receives the output of the P layer neuron, and the I layer neuron receives the output of a frame on the P layer neuron and performs local inhibition, which is specifically represented as:
Ef(x,y)=Pf(x,y); 2)
If(x,y)=∑i∑jPf-1(x+i,y+j)wj(i,j)(if i=j,j≠0); 3)
wherein: ef(x, y) is the excitatory membrane potential, i.e.the E-layer neuronal membrane potential; i isf(x, y) is the inhibitory membrane potential, i.e., the layer I neuronal membrane potential; w is aj(i, j) is the local suppression weight; i.e. iJ is not 0 at the same time;
s23: the method comprises the following steps that (1) the neurons in the S layer converge output in the corresponding position field of the neurons in the I layer, and the neurons in the E layer are inhibited through inhibiting membrane potential to obtain the membrane potential of the neurons in the S layer, and specifically the membrane potential of the neurons in the S layer is as follows:
Sf(x,y)=Ef(x,y)-If(x,y)WI; 5)
wherein: sf(x, y) is the membrane potential of S-layer neurons; wIIs a suppression weight matrix;
s24: the G-layer neuron is used for enhancing and extracting collision objects under the complex background, and the G-layer neuron membrane potential is obtained through calculation according to the S-layer neuron membrane potential, and the method specifically comprises the following steps:
wherein: gf(x, y) is G-layer neuronal membrane potential; [ w ]e]Is a convolution kernel; r is the convolution kernel radius, and r is 1;
setting a threshold TdeTo filter weak excitation points, specifically:
wherein:is a filtered G-layer neuronal membrane potential, CdeHas a coefficient of weakness of [0, 1%];TdeIs a filtering threshold;
s25: when the field image information reaches the LGMD neuron, the field image information is diagonally segmented to obtain coordinates on a y axis of two diagonal lines, which are specifically expressed as:
wherein: diag1, Diag2 are y coordinates of x corresponding to two diagonal lines, respectively; w is the width of the image; h is the height of the image; thereby obtaining the membrane potential of the 4-direction C-LGMD, which is specifically as follows:
wherein: u shapeLGMD,ULGMD,LLGMD,RLGMDMembrane potentials of 4 directions of upper, lower, left and right of the image respectively; adding the membrane potentials of the 4 orientations C-LGMD to obtain the LGMD neuron membrane potential KfThe method specifically comprises the following steps:
will KfNormalized and mapped to [0,255 ]]The range of (a) is specifically:
wherein n iscellThe total number of pixel points of the image;
obtaining the membrane potential after mapping in 4 directions according to the proportion of the membrane potential of the C-LGMD in the whole image in the 4 directions, which specifically comprises the following steps:
wherein the content of the first and second substances,respectively mapping the membrane potentials of the upper, lower, left and right 4 azimuths of the image; when K isfExceeds its threshold value TsThen an LGMD peak pulse is generatedThe method specifically comprises the following steps:
if n is continuoustsPulse is not less than nspThen it is judged that a collision is about to occur, specificallyExpressed as:
s26: the feedforward inhibition FFI neuron directly acquires field-of-view image information from the P layer neuron, which is specifically represented as:
wherein, FfInhibiting the membrane potential of an FFI neuron for feedforward; t isFFIIs a preset threshold value; when F is presentfExceeds a threshold value TFFIThe LGMD neuron membrane potential is immediately suppressed;
s27: selecting the direction: FFI membrane potential F is inhibited by current feedfGreater than a threshold value TFFIAll obstacle avoidance instructions are invalid; when C is presentfinalTurn and feed forward suppression of FFI membrane potential FfLess than threshold TFFIThen, the appearance of the obstacle is judged, and 4 azimuth films are comparedThe potential is the minimum position of the membrane potential in 4 positions as the safest obstacle avoidance direction;
s28: predicting a flight scene: when C is presentfinalFALSE and feed forward suppression of FFI Membrane potential FfLess than threshold TFFIWhen the unmanned aerial vehicle flies normally, predicting an unknown flying scene of the unmanned aerial vehicle by acquiring membrane potentials of 4 orientations of the FFI;
s29: and forming an obstacle avoidance control command by using the signals obtained in the steps S25, S27 and S28 as obstacle avoidance control signals.
6. The LGMD-based bionic unmanned aerial vehicle obstacle avoidance control method according to claim 5, wherein the method comprises the following steps: the process of predicting the flight scene in the step S28 specifically includes:
setting a scene prediction threshold TFFIIFFI average membrane potential when N frames are aheadLess than threshold TFFIIIn time, the unmanned aerial vehicle flies normally; FFI average membrane potential when N frames are aheadGreater than a threshold value TFFIIIs less than a threshold value TFFIBy comparisonFinding out the direction with the minimum membrane potential so as to predict the direction with the minimum future flight direction barrier in the front scene;
wherein, the FFI average membrane potential is calculated by N frames aheadThe concrete expression is as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910281845.5A CN109960278B (en) | 2019-04-09 | 2019-04-09 | LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910281845.5A CN109960278B (en) | 2019-04-09 | 2019-04-09 | LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109960278A CN109960278A (en) | 2019-07-02 |
CN109960278B true CN109960278B (en) | 2022-01-28 |
Family
ID=67025945
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910281845.5A Active CN109960278B (en) | 2019-04-09 | 2019-04-09 | LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109960278B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110908399B (en) * | 2019-12-02 | 2023-05-12 | 广东工业大学 | Unmanned aerial vehicle autonomous obstacle avoidance method and system based on lightweight neural network |
CN111831010A (en) * | 2020-07-15 | 2020-10-27 | 武汉大学 | Unmanned aerial vehicle obstacle avoidance flight method based on digital space slice |
CN114217621B (en) * | 2021-12-15 | 2023-07-07 | 中国科学院深圳先进技术研究院 | Robot collision sensing method and sensing system based on bionic insect vision |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010129907A2 (en) * | 2009-05-08 | 2010-11-11 | Scientific Systems Company Inc. | Method and system for visual collision detection and estimation |
CN107170289A (en) * | 2017-06-01 | 2017-09-15 | 岭南师范学院 | One kind is based on optics multi-vision visual vehicle rear-end collision early warning system |
US9984326B1 (en) * | 2015-04-06 | 2018-05-29 | Hrl Laboratories, Llc | Spiking neural network simulator for image and video processing |
CN108475058A (en) * | 2016-02-10 | 2018-08-31 | 赫尔实验室有限公司 | Time to contact estimation rapidly and reliably is realized so as to the system and method that carry out independent navigation for using vision and range-sensor data |
CN208126205U (en) * | 2018-04-28 | 2018-11-20 | 上海工程技术大学 | A kind of unmanned flight's device of automatic obstacle-avoiding |
-
2019
- 2019-04-09 CN CN201910281845.5A patent/CN109960278B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010129907A2 (en) * | 2009-05-08 | 2010-11-11 | Scientific Systems Company Inc. | Method and system for visual collision detection and estimation |
US9984326B1 (en) * | 2015-04-06 | 2018-05-29 | Hrl Laboratories, Llc | Spiking neural network simulator for image and video processing |
CN108475058A (en) * | 2016-02-10 | 2018-08-31 | 赫尔实验室有限公司 | Time to contact estimation rapidly and reliably is realized so as to the system and method that carry out independent navigation for using vision and range-sensor data |
CN107170289A (en) * | 2017-06-01 | 2017-09-15 | 岭南师范学院 | One kind is based on optics multi-vision visual vehicle rear-end collision early warning system |
CN208126205U (en) * | 2018-04-28 | 2018-11-20 | 上海工程技术大学 | A kind of unmanned flight's device of automatic obstacle-avoiding |
Non-Patent Citations (6)
Title |
---|
A Bio-inspired Collision Detector for Small Quadcopter;Jiannan Zhao;《 2018 International Joint Conference on Neural Networks (IJCNN)》;20180815;第1-7页 * |
A bio-inspired embedded vision system for autonomous micro-robots: the LGMD case;Cheng Hu;Cheng Hu;《 IEEE Transactions on Cognitive and Developmental Systems》;20160530;第9卷(第3期);第241-254页 * |
Collision detection in complex dynamic scenes using an LGMD-based visual neural network with feature enhancement,;Shigang Yue;《IEEE Transactions on Neural Networks》;20060508;第17卷(第3期);第705-716页 * |
Collision selective LGMDs neuron models research benefits from a vision-based autonomous micro robot;Qinbing Fu;《 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)》;20171214;第3996-4002页 * |
Obstacle avoidance with LGMD neuron: towards a neuromorphic UAV implementation;Llewyn Salt;《 2017 IEEE International Symposium on Circuits and Systems (ISCAS)》;20170928;第1-4页 * |
基于生物视觉的碰撞预警传感器;张国鹏;《传感器与微系统》;20160615;第35卷(第3期);第70-73页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109960278A (en) | 2019-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109144095B (en) | Embedded stereoscopic vision-based obstacle avoidance system for unmanned aerial vehicle | |
CN109034018B (en) | Low-altitude small unmanned aerial vehicle obstacle sensing method based on binocular vision | |
CN106681353B (en) | The unmanned plane barrier-avoiding method and system merged based on binocular vision with light stream | |
CN109960278B (en) | LGMD-based bionic obstacle avoidance control system and method for unmanned aerial vehicle | |
CN109460709A (en) | The method of RTG dysopia analyte detection based on the fusion of RGB and D information | |
CN111399505A (en) | Mobile robot obstacle avoidance method based on neural network | |
CN107817820A (en) | A kind of unmanned plane autonomous flight control method and system based on deep learning | |
CN104331901A (en) | TLD-based multi-view target tracking device and method | |
CN111461048B (en) | Vision-based parking lot drivable area detection and local map construction method | |
CN111186379B (en) | Automobile blind area dangerous object alarm method based on deep learning | |
CN114266889A (en) | Image recognition method and device, readable medium and electronic equipment | |
CN114359714A (en) | Unmanned body obstacle avoidance method and device based on event camera and intelligent unmanned body | |
CN106155082B (en) | A kind of unmanned plane bionic intelligence barrier-avoiding method based on light stream | |
Liu et al. | A novel trail detection and scene understanding framework for a quadrotor UAV with monocular vision | |
CN110610130A (en) | Multi-sensor information fusion power transmission line robot navigation method and system | |
CN111611869B (en) | End-to-end monocular vision obstacle avoidance method based on serial deep neural network | |
CN112380933B (en) | Unmanned aerial vehicle target recognition method and device and unmanned aerial vehicle | |
WO2023155903A1 (en) | Systems and methods for generating road surface semantic segmentation map from sequence of point clouds | |
Samal et al. | Closed-loop approach to perception in autonomous system | |
US20220377973A1 (en) | Method and apparatus for modeling an environment proximate an autonomous system | |
Hu et al. | Coping with multiple visual motion cues under extremely constrained computation power of micro autonomous robots | |
Lv et al. | Target recognition algorithm based on optical sensor data fusion | |
Cai et al. | LWDNet-A lightweight water-obstacles detection network for unmanned surface vehicles | |
CN112731918B (en) | Ground unmanned platform autonomous following system based on deep learning detection tracking | |
CN114217621B (en) | Robot collision sensing method and sensing system based on bionic insect vision |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20231211 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Address before: 524037 No.29 Cunjin Road, Chikan District, Zhanjiang City, Guangdong Province Patentee before: LINGNAN NORMAL University |