CN106681353A - Unmanned aerial vehicle (UAV) obstacle avoidance method and system based on binocular vision and optical flow fusion - Google Patents

Abstract

The invention discloses an unmanned aerial vehicle (UAV) obstacle avoidance method and system based on binocular vision and optical flow fusion. The method includes the steps of obtaining image information through an airborne binocular camera in real time; obtaining image depth information by using the graphics processing unit (GPU); extracting geometric contour information of the most threatening obstacle by using the obtained depth information and calculating the threat distance by a threat depth model; obtaining an obstacle tracking window by the rectangular fitting of the geometric contour information of the obstacle and calculating the optical flow field of the area to which the obstacle belongs to obtain the velocity of the obstacle relative to the UAV; and sending out an avoidance flight action instruction by a flight control computer to avoid the obstacle according to the calculated obstacle distance information, the geometric contour information and the relative velocity information. The invention realizes effective fusion of the depth information of the obstacle and the optical flow vector, obtains the movement information of the obstacle relative to the UAV in real time, improves the ability of the UAV to realize the visual obstacle avoidance, and enables a greater increase in the real-time performance and accuracy as compared with traditional algorithms.

Description

Based on binocular vision and the unmanned plane barrier-avoiding method and system of light stream fusion

Technical field

The present invention relates to unmanned plane barrier-avoiding method, more particularly to based on binocular vision and the unmanned plane avoidance of light stream fusion Method and system, belong to unmanned plane avoidance technical field.

Background technology

With the development of unmanned air vehicle technique and its application market, unmanned plane is often faced with different from the past special Business, these tasks are put forward higher requirement to its quick identification, the ability for hiding barrier on traveling air route.View-based access control model Obstacle avoidance system possesses the spies such as equipment simple, low cost, good economy performance, applied range because generally adopting passive working method Point.

Compared to the obstacle avoidance system based on active sensors such as ultrasound wave, laser radars, vision obstacle avoidance system response speed Faster, precision is higher, can provide the such as more abundant information of color, texture, geometry, therefore has obtained more and more Concern.

Binocular vision is compared with monocular vision, it is possible to obtain the range information vertical with photographic head, significantly more efficient can sentence Break and barrier with the relative position of unmanned plane, it helps fast and accurately split barrier from complex background； At present binocular vision has been widely used in the multiple fields such as robot navigation, target following.

Optical flow method is a kind of method moved come prediction pixel point using the dependency of pixel intensity data in image sequence, Study brightness of image change in time to set up the sports ground of object pixel point set；Generally, light stream can be with The associated movement of target motion or both in camera motion, scene is carried out effectively and accurately measuring, its premeasuring can be with table Show the instantaneous velocity that target is moved.

Generally there are two kinds of sparse optical flow method and dense optical flow method to the computational methods of optical flow field.Sparse optical flow chooses some figures Characteristic point in image field scape, by the speed that whole sports ground is fitted to the measurement of these feature spot speed.Dense optical flow is then It is the sports ground for calculating whole region, movement velocity of the target area relative to camera is obtained with this；Sparse optical flow computing speed Degree is fast, but value of calculation error is big；Although dense optical flow calculating is more accurate, if not carrying out essence to the target area to be calculated Really segmentation, then can greatly increase the calculating time, so being often required to coordinate quick accurate image segmentation algorithm to use.

Application No. CN201410565278.3《Based on binocular stereo vision and the vehicle movement information inspection of light stream fusion Survey method》, mainly point-of-interest on ground is marked by binocular vision, then the light flow valuve of point-of-interest is calculated, again finally Estimate the D translation speed and three-dimensional rotation speed of surface car with least square fitting.The method adopts the speed of characteristic point Information replaces the velocity information of vehicle, although arithmetic speed is strengthened, but estimated accuracy is difficult to ensure that.And the method is right The movable information of vehicle itself is estimated, it is impossible to recognize and hide the barrier in traveling process, it is difficult to obtain in practical field To utilization.

Application No. CN201110412394.8《A kind of inspection prober automatic obstacle avoiding based on binocular stereo vision is advised The method of drawing》, mainly the three-dimensional coordinate of all pixels point is calculated in camera image by binocular stereo vision and forms sensing point Three-dimensional map, the optimal path of an avoiding obstacles is selected according to three-dimensional map.The method needs to calculate in visual field The three-dimensional coordinate of whole pixels, has very high requirement to the capacity of processor operational performance and memorizer, do not apply to it is little The embedded airborne equipment of type.

Apply as CN201510688485.2's《A kind of autonomous obstacle detection system of the unmanned plane based on binocular vision and Method》, having highlighted carries out the hardware structure of detection of obstacles using binocular camera, and it is mainly using FPGA as process The arithmetic core of binocular image.The features such as although FPGA has small volume, fast operation, but FPGA is expensive and needs to make Programming is carried out with special development language, is unfavorable for being docked with other modules.And the patent only illustrates unmanned plane The hardware structure of detection of obstacles is carried out using binocular vision, the specific algorithm to how to detect barrier is not illustrated.

Therefore, although have more research in the field that avoidance is carried out using binocular vision both at home and abroad, but big multi-method without Position and speed of the quick acquired disturbance thing of method relative to unmanned plane, it is difficult to rapidly and accurately barrier is hidden, therefore greatly It is difficult to apply to unmanned plane Real Time Obstacle Avoiding field more.

The content of the invention

The technical problem to be solved is：The unmanned plane barrier-avoiding method with light stream fusion based on binocular vision is provided And system, the depth information of barrier and light stream vector are carried out into effective integration, real-time acquired disturbance thing is relative to unmanned plane Movable information, realizes the real-time Obstacle avoidance of unmanned plane.

The present invention is employed the following technical solutions to solve above-mentioned technical problem：

Based on binocular vision and the unmanned plane barrier-avoiding method of light stream fusion, comprise the steps：

Step 1, using binocular camera the image in unmanned plane direction of advance is obtained, and does greyscale transformation to image；

Step 2, calculates after greyscale transformation on image the characteristic information of each pixel and carries out Stereo matching, obtains unmanned plane Depth map information in direction of advance；

Step 3, the depth value in depth map is divided into the class of depth value two for belonging to barrier or belonging to background, by depth map It is divided into barrier region and background area, using the maximum profile of closed area in barrier region as barrier profile, is used in combination Rectangle frame is fitted to it, obtains barrier and follows the trail of window as the geological information of barrier；

Step 4, using dense optical flow method the velocity that barrier follows the trail of window sliding is calculated, and obtains the window in x, y Speed on direction simultaneously follows the trail of the position of window, the position that will determine that and next frame according to the next two field picture barrier of velocity estimated The barrier that Practical Calculation goes out is followed the trail of the window's position and is compared, if difference between the two is less than threshold value, carries out step 5, Otherwise, return to step 1 is recalculated；

Step 5, barrier is calculated for the threat depth value of unmanned plane using depth threat modeling；

Step 6, the barrier velocity information that the barrier geological information that step 3 is calculated is calculated with step 4 is from picture Plain coordinate transformation is world coordinates, and the kinematic parameter using unmanned plane is corrected；

Step 7, the geometry that obstacle position information and step 6 are obtained, velocity information are sent to the winged control meter of unmanned plane In calculation machine, flight control computer controls unmanned plane and makes real-time avoiding action according to above- mentioned information.

As a kind of preferred version of the inventive method, unmanned plane advance side is obtained using binocular camera described in step 1 The detailed process of image upwards is：Binocular camera is installed on into the head position of unmanned plane, obtains double by scaling method The inside and outside parameter matrix of lens camera and distortion parameter, using binocular camera the image in unmanned plane direction of advance is obtained, and According to inside and outside parameter matrix and distortion parameter to correct image, undistorted and row alignment two width images are obtained.

Used as a kind of preferred version of the inventive method, the detailed process of the step 2 is：Calculate image after greyscale transformation Upper each pixel upper and lower, left and right, upper left, lower-left, upper right, the energy function in 8 directions in bottom right are simultaneously added up, and seek parallax value So that the energy function after cumulative is minimized, the depth value information of each pixel is determined according to parallax value.

As a kind of preferred version of the inventive method, by wheel that closed area in barrier region is maximum described in step 3 Before exterior feature is as barrier profile, the depth map for having distinguished barrier region and background area is entered using speckle wave filter Row filtering, removes noise.

As a kind of preferred version of the inventive method, the depth value in depth map is divided into described in step 3 belongs to obstacle The detailed process of thing or the class of depth value two for belonging to background is：

Setting segmentation threshold D_h, depth value is more than or equal to into D_hClassify as the depth value for belonging to barrier, depth value is little In D_hClassify as the depth value for belonging to background, by maximizing the square solution D between barrier and background_h, varianceMeter Calculating formula is：

Wherein, ω₀、ω₁Respectively depth value is by D_hIt is divided into barrier, the probability of background depth value, μ₀、μ₁Respectively belong to In barrier, the average of the depth value of background：

Wherein, D_iFor discrete credible depth layer, i=1 ..., t, t for depth layer number, D₁,…,D_hTo belong to obstacle The depth layer of thing, D_h+1,…,D_tTo belong to the depth layer of background, K_jFor the number of depth value in each depth layer, j=1 ..., t.

Used as a kind of preferred version of the inventive method, the detailed process of the step 5 is：

The depth value set for belonging to barrier is set as D_K={ d₁,d₂,…,d_K, d₁,d₂,…,d_KDepth value is, K is Belong to the number of the depth value of barrier；D₁,…,D_tFor discrete credible depth layer, t for depth layer number, K₁,…,K_tFor The number of depth value in each depth layer, if having at least one1≤j≤t, then barrier is for the threat of unmanned plane Depth value is：Wherein, D_minBe more thanK_jIt is minimum in corresponding depth layer Depth layer, K_minFor D_minThe number of middle depth value；If all of K₁,…,K_tRespectively less thanThen barrier is for the prestige of unmanned plane Coercing depth value is：

Based on binocular vision and the unmanned plane obstacle avoidance system of light stream fusion, including the image acquisition mould being mounted on unmanned plane Block, image processing module, inertia measuring module, GNSS module, unmanned plane includes flight control computer；Described image acquisition module is obtained The incoming image processing module of image synchronization in unmanned plane direction of advance is taken, image processing module includes CPU module and GPU moulds Block, calculates respectively geometry, speed, the positional information of barrier, and GNSS module, inertia measuring module carry out reality to unmanned plane respectively Shi Dingwei and attitude measurement, inertia measuring module is also corrected to the velocity information that image processing module is calculated, and flies control meter Calculation machine according to the information that image processing module sends merged and controlled unmanned plane make avoid before it is dynamic to evading for barrier Make.

Used as a kind of preferred version of present system, the system also includes ultrasonic wave module, and ultrasonic wave module includes four Individual ultrasonic sensor, is respectively arranged on the front, rear, left and right four direction of unmanned plane, for detecting the obstacle of four direction Thing.

The present invention adopts above technical scheme compared with prior art, with following technique effect：

1st, the present invention is effectively split by binocular vision to barrier, only calculates the light stream that barrier is followed the trail of in window Value, solve the problems, such as dense optical flow calculate entire image when time-consuming, further increase the real-time of obstacle avoidance algorithm.

2nd, the present invention proposes a kind of new threat depth model, and the model can simplify avoidance flow process, it is not necessary to introduce excessively Complicated path planning algorithm, with higher practical value.

3rd, the present invention calculates Stereo matching and light stream by being calculated using GPU and CPU simultaneously using GPU, utilizes CPU calculates barrier physical dimension, position and speed, improves the arithmetic speed of algorithm.

Description of the drawings

Fig. 1 is the chessboard trrellis diagram that photographic head is demarcated in the present invention.

Fig. 2 is the hardware structure diagram of the unmanned plane obstacle avoidance system that the present invention is merged based on binocular vision with light stream.

Fig. 3 is the algorithm flow chart of the unmanned plane barrier-avoiding method that the present invention is merged based on binocular vision with light stream.

Fig. 4 is for the Robot dodge strategy schematic diagram of the continuous barrier of depth in the embodiment of the present invention.

Specific embodiment

Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the drawings.Below by The embodiment being described with reference to the drawings is exemplary, is only used for explaining the present invention, and is not construed as limiting the claims.

As shown in Fig. 2 the hardware structure diagram of the unmanned plane obstacle avoidance system merged with light stream based on binocular vision for the present invention, System includes image capture module, embedded image processing module, flight control computer, ultrasonic wave module, GNSS (Global Navigation Satellite System GPSs) module and inertia measuring module, wherein, embedded figure As processing module includes CPU module and CPU module, ultrasonic wave module includes four ultrasonic sensors, before unmanned plane, Afterwards, on left and right four direction mounting ultrasonic sensor as obstacle avoidance aiding device, when ultrasonic sensor sensor is detected When other directions have barrier, flight control computer adjusts avoidance action according to obstacle distance information.

Image capture module is by the synchronous incoming embedded image processing module of left images；CPU module and CPU module are responsible for Calculate the avoidance parameter of unmanned plane；GNSS module and inertia measuring module are responsible for the real-time positioning and attitude measurement of unmanned plane, this Outward, inertia measuring module is also responsible for by serial ports measurement result being sent to embedded image processing module with to the speed for calculating Degree information is corrected；Flight control computer carries out embedded image processing module and ultrasonic wave module transmitted information effectively Merge and control unmanned plane make avoid before to barrier avoiding action.

Image capture module adopts binocular camera, resolution 640*480 or 800*600 etc. in this example, can use frame per second 20 arrive 30fps, parallax range 12cm, and parallax range is adjustable with focal length, can be by USB or other high-speed interfaces directly by two-way Synchronous video signal sends into embedded image processing module.

As shown in figure 3, the present invention is as follows with the algorithm flow of the unmanned plane barrier-avoiding method that light stream is merged based on binocular vision： Binocular camera is demarcated first by gridiron pattern as shown in Figure 1, respectively obtains the Intrinsic Matrix of two video cameras And the outer parameter (including spin matrix and translation vector) between distortion parameter, two video cameras and it is stored in embedded figure As in the memorizer of processing module.

The synchronizing video data that binocular camera sends is read in, using Intrinsic Matrix, distortion parameter and outer parameter by two Width image is corrected, and makes two width images undistorted and row alignment；Two width images are transformed to into gray space and CPU module is utilized Simultaneously each pixel upper and lower, left and right are calculated, upper left, lower-left, upper right, the energy function in 8 directions in bottom right are simultaneously accumulated it, By calculate each pixel simultaneously so that the parallax value that energy function is minimized determine each pixel depth information (i.e. Depth map).CPU module obtains barrier and follows the trail of window according to the depth information that CPU module is calculated, and determines the several of barrier What, positional information, and determine that barrier follows the trail of window；CPU module follows the trail of window parallel computation window further according to selected barrier To calculate relative velocity, last CPU module is corrected velocity information and and position letter to the light flow valuve of all pixels point in mouthful Breath, physical dimension send to flight control computer processed together.

Depth value in depth map is divided into into two classes, a class is the depth value D for belonging to barrier_K={ d₁,d₂,…,d_K, One class is the depth value for belonging to backgroundBy maximizing D_KWith D_bBetween variance find segmentation threshold Value D_h, and will be less than D_hDepth value be set to 0, calculating profile in depth map and chooses largest contours and carries out rectangle after processing Fitting, obtains barrier rectangle and follows the trail of window.

Variance between two class depth valuesIt is represented by：

ω in formula₀And ω₁It is by D_hThe probability of two class depth values of segmentation, μ₀And μ₁For two class depth averages：

Wherein, { D₁,…,D_t}=D_rFor discrete credible depth layer, K_jFor the number of depth value in each depth layer, t is deep The number of degree layer.

Extract through barrier, the depth map of background differentiation and calculate the profile in depth map；Carry out contour detecting it Before, first depth map is filtered using speckle wave filter, remove less block depth areas；By the closure for detecting The maximum profile of area is fitted with rectangle frame as barrier profile to it, using the rectangle for fitting as barrier Follow the trail of window.

Calculate threat depth D of the barrier to unmanned plane₀, it is assumed that D_K={ d₁,d₂,…,d_KBe one group and belong to barrier Depth value set, d₁,d₂,…,d_KIt is belonging to the depth value of barrier, K is the number of these depth values, D_rFor believable depth Interval, D₁,D₂,…,D_tIt is discrete credible depth layer on this interval, { D₁,D₂,…,D_t}=D_r, K₁,…,K_tFor each depth Degree layer D₁,D₂,…,D_tThe number of middle depth value, as 1≤j≤t, if there isBarrier follows the trail of the threat depth of window Degree D₀For：

In formula, D_minBe more thanK_jMinimum-depth layer, K in corresponding depth layer_minFor D_minThe number of middle depth value；

If all ofThreat depth D of barrier tracking window₀For：

The optical flow field in barrier tracking window is calculated using Horn-Schunck dense optical flows method, tracking window is obtained and is existed Speed on x, y direction simultaneously follows the trail of the position of window according to velocity estimated next frame；The position that will determine that and the actual meter of next frame The tracking the window's position for calculating is compared, if difference between the two is less than 10 pixels, is judged to calculate accurate, if More than 10 pixels, then judge mistake in computation and return the first step to recalculate.

The barrier for calculating is followed the trail of the barrier geological information and barrier represented by window using inside and outside parameter matrix Hinder thing velocity information to be converted into world coordinates from pixel coordinate, and using below equation to barrier relative to unmanned plane speed It is corrected：

In formula, v_xAnd v_yFor the unmanned plane x after correction, y directions speed；f_xAnd f_yFor x, the focal length on y directions；U and v are The barrier x calculated in step 4, y directions light stream vector；D₀For the threat depth of barrier tracking window；θ、Respectively The pitching of unmanned plane, yaw angle；Δ t is the time between two frames.

The speed for calculating, geometry, positional information and threat depth information are sent to into UAV Flight Control System, nothing Man-machine flight control system makes real-time avoiding action to avoid the barrier on flight path according to the information control unmanned plane, Positional information is that barrier follows the trail of deviant of the window center relative to picture centre.

As shown in figure 4, when the continuous barrier of depth is hidden, multiple barriers are broken down into, in different two field pictures The middle threat depth for calculating barrier each section respectively carries out avoidance.

The present invention obtains depth map information, then break the barriers segmentation, prestige by the Stereo Matching Algorithm accelerated through GPU The relative position of side of body depth calculation, the physical dimension of the step acquired disturbance thing such as optical flow computation and unmanned plane, speed and by its Deliver to flight control computer and produce avoidance action.The algorithm improves the fortune of algorithm by being calculated with CPU using GPU simultaneously Speed is calculated, barrier is effectively split by binocular vision, only calculate the light flow valuve that barrier is followed the trail of in window, solved The problem that time-consuming during dense optical flow calculating entire image, further increases the real-time of obstacle avoidance algorithm；It is proposed by the present invention Depth model is threatened to simplify avoidance flow process, it is not necessary to excessively complicated path planning algorithm to be introduced, with higher practical value.

Above example technological thought only to illustrate the invention, it is impossible to which protection scope of the present invention is limited with this, it is every According to technological thought proposed by the present invention, any change done on the basis of technical scheme, the scope of the present invention is each fallen within Within.

Claims (8)

1. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision, it is characterised in that comprise the steps：

Step 1, using binocular camera the image in unmanned plane direction of advance is obtained, and does greyscale transformation to image；

Step 2, calculates after greyscale transformation on image the characteristic information of each pixel and carries out Stereo matching, obtains unmanned plane advance Depth map information on direction；

Step 3, the depth value in depth map is divided into the class of depth value two for belonging to barrier or belonging to background, and depth map is divided into Barrier region and background area, using the maximum profile of closed area in barrier region as barrier profile, and use rectangle Frame is fitted to it, obtains barrier and follows the trail of window as the geological information of barrier；

Step 4, using dense optical flow method the velocity that barrier follows the trail of window sliding is calculated, and obtains the window in x, y directions On speed and the position of window, the position that will determine that and next frame reality are followed the trail of according to the next two field picture barrier of velocity estimated The barrier for calculating is followed the trail of the window's position and is compared, if difference between the two is less than threshold value, carries out step 5, otherwise, Return to step 1 is recalculated；

Step 5, barrier is calculated for the threat depth value of unmanned plane using depth threat modeling；

Step 6, the barrier velocity information that the barrier geological information that step 3 is calculated is calculated with step 4 is sat from pixel Mark is converted into world coordinates, and the kinematic parameter using unmanned plane is corrected；

Step 7, the geometry that obstacle position information and step 6 are obtained, velocity information are sent to the flight control computer of unmanned plane In, flight control computer controls unmanned plane and makes real-time avoiding action according to above- mentioned information.

2. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision according to claim 1, it is characterised in that step It is using the detailed process of the image in binocular camera acquisition unmanned plane direction of advance described in 1：Binocular camera is installed on The head position of unmanned plane, obtains the inside and outside parameter matrix and distortion parameter of binocular camera, using binocular by scaling method Video camera obtains the image in unmanned plane direction of advance, and according to inside and outside parameter matrix and distortion parameter to correct image, Obtain undistorted and row alignment two width images.

3. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision according to claim 1, it is characterised in that described The detailed process of step 2 is：Calculate after greyscale transformation each pixel upper and lower, left and right, upper left, lower-left, upper right, bottom right on image The energy function in 8 directions is simultaneously added up, and asks parallax value that the energy function after adding up is minimized, and is determined according to parallax value The depth value information of each pixel.

4. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision according to claim 1, it is characterised in that step Described in 3 using the maximum profile of closed area in barrier region as before barrier profile, using speckle wave filter to area Divide barrier region and the depth map of background area to be filtered, remove noise.

5. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision according to claim 1, it is characterised in that step The detailed process that depth value in depth map is divided into the class of depth value two for belonging to barrier or belonging to background described in 3 is：

Setting segmentation threshold D_h, depth value is more than or equal to into D_hClassify as the depth value for belonging to barrier, by depth value be less than D_h Classify as the depth value for belonging to background, by maximizing the square solution D between barrier and background_h, varianceCalculate public Formula is：

${\mathrm{\σ}}_d^2={\mathrm{\ω}}_0{({\mathrm{\μ}}_0-{\mathrm{\μ}}_1)}^2+{\mathrm{\ω}}_1{({\mathrm{\μ}}_0-{\mathrm{\μ}}_1)}^2,$

Wherein, ω₀、ω₁Respectively depth value is by D_hIt is divided into barrier, the probability of background depth value, μ₀、μ₁Respectively belong to barrier Hinder thing, the average of the depth value of background：

$\begin{array}{cc}{\mathrm{\ω}}_0=\underset{i=1}{\overset{h}{\Σ}}(\underset{j=1}{\overset{i}{\Σ}}{K}_j/t)& {\mathrm{\μ}}_0=\frac{\underset{i=1}{\overset{h}{\Σ}}{D}_i\frac{K_i}{t}}{{\mathrm{\ω}}_0}\end{array}$

$\begin{array}{cc}{\mathrm{\ω}}_1=\underset{i=h+1}{\overset{t}{\Σ}}(\underset{j=1}{\overset{i}{\Σ}}{K}_j/t)& {\mathrm{\μ}}_1=\frac{\underset{i=h+1}{\overset{t}{\Σ}}{D}_i\frac{K_i}{t}-\underset{i=i}{\overset{h}{\Σ}}{D}_i\frac{K_i}{t}}{1-{\mathrm{\ω}}_0}\end{array}$

Wherein, D_iFor discrete credible depth layer, i=1 ..., t, t for depth layer number, D₁,…,D_hTo belong to barrier Depth layer, D_h+1,…,D_tTo belong to the depth layer of background, K_jFor the number of depth value in each depth layer, j=1 ..., t.

6. the unmanned plane barrier-avoiding method for being merged with light stream based on binocular vision according to claim 1, it is characterised in that described The detailed process of step 5 is：

The depth value set for belonging to barrier is set as D_K={ d₁,d₂,…,d_K, d₁,d₂,…,d_KDepth value is, K is to belong to The number of the depth value of barrier；D₁,…,D_tFor discrete credible depth layer, t for depth layer number, K₁,…,K_tFor each depth The number of depth value in degree layer, if having at least one1≤j≤t, then barrier is for the threat depth of unmanned plane It is worth and is：Wherein, D_minBe more thanK_jMinimum-depth in corresponding depth layer Layer, K_minFor D_minThe number of middle depth value；If all of K₁,…,K_tRespectively less thanThen barrier is for the threat depth of unmanned plane Angle value is：

7. the unmanned plane obstacle avoidance system based on binocular vision and light stream fusion, it is characterised in that including being mounted on unmanned plane Image capture module, image processing module, inertia measuring module, GNSS module, unmanned plane includes flight control computer；Described image Acquisition module obtains the incoming image processing module of image synchronization in unmanned plane direction of advance, and image processing module includes CPU moulds Block and CPU module, calculate respectively geometry, speed, the positional information of barrier, and GNSS module, inertia measuring module are respectively to nothing Man-machine to carry out real-time positioning and attitude measurement, the velocity information that inertia measuring module is also calculated to image processing module carries out school Just, flight control computer according to the information that image processing module sends merged and controlled unmanned plane make avoid before to barrier Avoiding action.

8. the unmanned plane obstacle avoidance system for being merged with light stream based on binocular vision according to claim 1, it is characterised in that this is System also include ultrasonic wave module, ultrasonic wave module include four ultrasonic sensors, be respectively arranged in unmanned plane it is forward and backward, left, On right four direction, for detecting the barrier of four direction.