CN105700525A - Robot working environment uncertainty map construction method based on Kinect sensor depth map - Google Patents

Abstract

A method for constructing an uncertainty map of a robot's working environment based on a Kinect sensor depth map: it is characterized in that: the method includes the following steps: Step (1): the robot uses the Kinect sensor to collect depth data; Step (2): the collected Preprocess the depth data to obtain the depth data map; step (3): collect the ground depth data and extract the ground model; step (4): cut the ground model to the depth data map to obtain the obstacle depth map, step (5 ): detect the obstacle depth map and identify the obstacle area, and step (6) analyze the obstacle area and the free area to form an uncertainty grid map of the robot's working environment. The invention can accurately detect the surrounding environment and establish an uncertain grid map, which provides preconditions and conditions for the robot to complete other tasks such as obstacle avoidance, navigation, and path planning.

Description

Construction Method of Uncertainty Map of Working Environment of Robot Based on Kinect Sensor Depth Map

Technical field: the present invention relates to a kind of robot working environment uncertainty map construction method based on Kinect sensor depth map. The invention realizes that the robot can detect the surrounding environment and form an uncertainty map, which can provide preconditions and conditions for the robot to complete other tasks such as obstacle avoidance, navigation, and path planning.

Background technology: Constructing a map is one of the core contents of mobile robotics research. Its purpose is to better represent the surrounding environmental information through the construction of the map, which is more conducive to the robot's identification of environmental information for subsequent work. At present, there are many methods for building an environmental map for robots. The environmental map construction method based on laser sensors has the disadvantages of high sensor prices and low cost performance; the environmental map construction method based on ultrasonic sensors has relatively rough environmental information and low accuracy. and other disadvantages; the environment map construction method based on visual sensors has the disadvantages of complex calculation and difficult implementation. The sensor used in the present invention is Kinect, which is a new sensor released by Microsoft Corporation in 2010. It can not only obtain the optical image of the environment but also obtain the position information of the object on the optical image. It has good adaptability, simple structure, strong real-time performance and low price, so it can become a favorable tool for robot environment perception. The Kinect sensor collects 3D information of the indoor environment through a color camera and a depth camera, and determines the color information and depth information of each point in the environment by outputting an RGB image and an infrared depth image. The map established by the present invention is a grid map. A grid map discretizes the environment into a series of grids, each of which has a state. The Kinect sensor has the characteristic that the detected depth distance error will become larger with the increase of the distance, so the detected obstacles in the grid also have uncertainty. The resulting map is an uncertainty grid map.

Invention content:

Purpose of the invention: the present invention provides a method for constructing a robot working environment uncertainty map based on a depth map of a Kinect sensor, the purpose of which is to realize the detection of the surrounding environment and construct a map for the follow-up work of the robot.

Technical solution: the present invention is implemented through the following technical solutions:

1. A method for constructing a robot working environment uncertainty map based on a Kinect sensor depth map: it is characterized in that: the method includes the following steps:

Step (1): the robot uses the Kinect sensor to collect depth data;

Step (2): Preprocessing the collected depth data to obtain a depth data map;

Step (3): collecting ground depth data and extracting the ground model;

Step (4): Carry out ground model shearing processing on the depth data map to obtain the obstacle depth map, and then perform shearing processing on the original depth data map and the obstacle depth map to obtain the ground depth map;

Step (5): Detect the obstacle depth map and identify the obstacle area, detect the ground depth map and identify the free area, analyze the obstacle area and the free area to form an uncertainty grid map;

Step (6) Analyze the obstacle area and the free area to form an uncertainty grid map of the robot's working environment.

The method that described step (3) adopts for the extraction of ground model is to collect the depth map under the environment of an open space without obstacles, according to the imaging principle of Kinect, it can be learned that the depth image has the following properties: (1) and the feature of the image Nothing to do, only distance. (2) The change direction of the gray value is consistent with the z-axis direction of the field of view captured by the Kinect depth camera, and the gray value will become larger as the distance increases. So the depth information of the ground detected at the same distance as the Kinect is the same. Set the relative height and pitch angle of the Kinect sensor to the ground to be constant, and collect a depth map in an open and unobstructed environment. When the distance exceeds a certain threshold, Kinect can no longer detect the ground data ahead, so only Kinect is used. The ground data that can be detected is regarded as invalid data in other places and recorded as 0; due to the performance limitation of the Kinect sensor itself, it is better for collecting ground information near the ground, and it is worse in the distance, and the data collected in the closer place The more complete the data is, but the data collected in the distance is incomplete and has large errors, so it needs to be processed; each line of the depth image records the ground depth information at the same distance as Kinect, and it is scanned to record the depth information of each line. Depth information, remove the depth information of invalid data, and the final data obtained by weighted average of the remaining data is the ground depth information of the place; process the data of each row and record and generate a ground model template; thus we can get A ground model; save the data in the ground model in the root directory of the program.

The area of the map in the step (5) is divided into a free area, an obstacle area and an unknown area. The free area is set as the detected ground area, the obstacle area is set as the detected area with obstacles, and the unknown area is set as other areas except the ground and obstacles. Use a structure to record raster information. These include the status identification of the grid, the confidence of the grid, and the color of the grid. The specific operation includes the following steps:

(1) Idle area detection algorithm: compare the depth data collected by Kinect with the obtained ground depth data, if the difference between the ground depth data and the collected depth data is less than a certain threshold, the ground depth data will be saved, otherwise the data will be set to is 0. The obtained ground depth information is mapped to the world coordinate system and the occupied grid information is recorded.

(2) Obstacle area detection algorithm: compare the depth data collected by Kinect with the obtained ground depth data, if the difference between the ground depth data and the collected depth data is less than a certain threshold, set the data to 0, otherwise keep the collection to the depth data. Obtained is the obstacle depth information, which is mapped to the world coordinate system after scanning and analyzing the depth data series and records the occupied grid information.

The formula for determining the obstacle confidence is obtained through the analysis of the characteristics of the depth data collected by the Kinect sensor and the obstacle confidence determination model. Get the uncertainty grid map.

Advantages and effects:

The invention uses the Kinect sensor to realize the construction of the local grid map, and divides the environment into three parts: idle area, obstacle area and unknown area; the robot can move in the idle area, but cannot move in the obstacle area, and the corresponding unknown area needs to be re-detected. Compared with the visual sensor, the present invention can not only obtain the color information of the environment but also obtain the distance information, and can build a better map; compared with the ultrasonic sensor, the environmental information obtained by the present invention is more refined and has higher precision; compared with the laser sensor In comparison, the detection range of the present invention is larger, and three-dimensional information can be obtained, and the cost performance is also higher.

In the present invention, the ground model shearing process is performed on the depth map collected by the Kinect sensor to remove the influence of the ground on obstacle detection, and the rapid detection of obstacles is realized through the column scanning method of the obstacle depth map; according to the Kinect Due to the limitations of the sensor itself, the obstacle grid confidence model is established to determine the grid obstacle confidence, which realizes the uncertainty establishment of the grid and makes the establishment of the map more accurate. Finally, the present invention can accurately detect the surrounding environment and establish an uncertainty grid map, which provides prerequisites and conditions for the robot to complete other tasks such as obstacle avoidance, navigation, and path planning.

Description of drawings:

Figure 1 is the original ground depth map;

Figure 2 is the ground depth map after processing;

Figure 3 is the original depth map;

Figure 4 is the obstacle depth map after cutting the ground model;

Figure 5 is the ground depth map after cutting obstacles

Figure 6 is the obstacle distribution coordinate system

Figure 7 is the uncertainty grid map

Figure 8 is the obstacle confidence determination model

The specific embodiment: the present invention is specifically described below in conjunction with accompanying drawing:

A kind of robot working environment uncertainty map construction method based on Kinect sensor depth map of the present invention comprises the following steps:

Step 1: The robot uses the Kinect sensor to collect depth data. This step uses the Kinect sensor to store the collected depth data in a one-dimensional array for subsequent processing.

Step 2: Preprocessing the collected depth data to obtain a depth data map. The present invention firstly needs to map the depth information to the color information so as to facilitate the image display. Through the experimental test, it is found that the effective distance that Kinect can detect is within 10 meters, so 0 to 10 meters are mapped to 0 to 255, that is, the distance Mapped to color to enable the display of the depth map. Get the depth information color map. As shown in Figure 3.

Step 3: Collect ground depth data and extract the ground model. Since the depth information collected by Kinect is only related to distance, the depth information of the ground at the same distance from Kinect should be the same. Then the ground information can be extracted as a template.

In this step, the method that the present invention adopts is to gather depth map under the environment of an open space without obstacle, according to the imaging principle of Kinect, can know that depth image has the following properties: (1) has nothing to do with the feature of image, only with It's about distance. (2) The change direction of the gray value is consistent with the z-axis direction of the field of view captured by the Kinect depth camera, and the gray value will become larger as the distance increases. So the depth information of the ground detected at the same distance as the Kinect is the same. Set the relative height and pitch angle of the Kinect sensor to the ground to be constant, and collect a depth map in an open and unobstructed environment. When the distance exceeds a certain threshold, Kinect can no longer detect the ground data in front, so only Kinect is used. The ground data that can be detected are regarded as invalid data in other places and recorded as 0. Due to the limitation of the performance of the Kinect sensor itself, it is relatively good at collecting ground information near the ground, and relatively poor at far away. It also needs to be processed. Each line of the depth image records the depth information of the ground at the same distance as Kinect, and performs line scanning to record the depth information of each line, removes the depth information detected as 0, and weights the remaining data to obtain the final data It is the ground depth information at this place. Process each line of data and record and generate a ground model template, so that a ground model can be obtained. The data in the ground model is saved in the program root directory. The original ground depth map is shown in Figure 1, and the processed ground model depth map is shown in Figure 2.

Step 4: Obtain the ground depth map by clipping the obstacles on the depth map, detect the obstacle depth map and identify the obstacle area, detect the ground depth map and identify the free area, and form uncertainty in the analysis of the obstacle area and the free area grid map.

Obstacle area detection algorithm:

The specific implementation steps of the obstacle area detection algorithm are as follows:

(1) Compare the depth data collected by Kinect with the obtained ground depth data, if the difference between the ground depth data and the collected depth data is less than a certain threshold, set the data to 0, otherwise keep the collected depth data. Obtained obstacle depth data shown in Figure 4.

(2) Use the scanning method to perform column scanning on the obtained depth map, taking the first column as an example: when scanning to the first non-zero number, record this number as the seed point of the first obstacle, and when scanning to the second When two non-zero numbers are compared with the first one, if the difference between the two is less than a certain threshold, the two are merged into one seed point, and the average value of the two is taken as the new seed point. If the difference between the two exceeds a certain threshold, record the latter as a new seed point. until a column is scanned. A structure is used to record the obstacle information of each column, including the number of obstacles, the distance of obstacles, the number of pixels contained in obstacles, the coordinates of the top of obstacles, and the coordinates of bottom of obstacles.

(3) Repeat step 2 continuously to obtain various information of all different obstacles in all columns, judge different obstacles, and remove all obstacles whose number of pixels contained in the obstacles is less than a certain threshold.

(4) According to step 3, a coordinate system in which the abscissa is the pixel position of the image and the ordinate is the actual distance can be obtained. Each point in the coordinate system represents an obstacle. The result is shown in Figure 6.

(5) The coordinate system obtained according to step 4 is converted into the obstacle display under the actual distance coordinates again; it is necessary to convert the image coordinate system to the camera coordinate system and then to the world coordinate system; use the formula (1) to obtain the obstacle data from the image The coordinate system is converted to coordinates in the world coordinate system;

$dx=|v-v_0|\frac{depth(u,v)}{f_x}$ Formula 1)

dz=depth(u,v)

In the above formula, dx represents the offset distance of the pixel point (u, v) relative to the center position (u ₀ , v ₀ ) in the X direction, dz represents the depth distance corresponding to the point, and f _x is the internal parameter of the camera representing X The focal length in the direction is set to a constant value;

(6) Determine which grid the obstacle data belongs to in the world coordinate system, and record the grid;

Free area detection algorithm:

The specific implementation steps of the free area detection algorithm are as follows:

(1) Compare the depth data collected by Kinect with the obtained ground depth data, if the difference between the ground depth data and the collected depth data is less than a certain threshold, the ground depth data is saved, otherwise the data is set to 0. The obtained ground depth data map is shown in Fig. 5.

(2) Use the formula (1) to obtain the ground data from the image coordinate system to the coordinates in the world coordinate system;

(3) Determine which grid the ground data belongs to in the world coordinate system, and record the grid.

Step 5: Confidence analysis is performed on the obstacle to determine the confidence of the obstacle.

The specific implementation steps of the obstacle confidence confirmation algorithm are as follows:

Because Kinect has errors in distance measurement, it is necessary to perform confidence analysis on the grid. As the depth data detected by Kinect increases with the distance, the error also increases. And there is a certain ratio between the two, such as formula (2).

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; z</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; d</span>}$

In the above formula, σz represents the error at a distance of z, f represents the focal length of the depth camera, b represents the baseline length (the distance between the infrared transmitter and the receiver), m represents the normalized parameter, z represents the actual depth distance, and σd represents the two One-third of the pixel distance.

The obstacle confidence determination model is shown in Figure 8:

Obtain the obstacle confidence determination model; if an obstacle is detected on the grid, the probability of falling on the grid is (grid_length-2*σz) ² /grid_length ² . The formula (3) for calculating the grid confidence can be obtained.

$\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; R</span>}}{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; z</span>}}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}$

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; =</span>\frac{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; \_</span>{\mathrm{<span\; class="notranslate">\; length</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span>{{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&sigma;</span>}\mathrm{<span\; class="notranslate">\; d</span>}$

In the above formula, p represents the confidence of the grid, and f ₁ and f ₂ represent the proportions of the two types of data that affect the confidence. E represents the influence of error on grid confidence, R represents the influence of grid length on grid confidence, σMax represents the maximum error at the furthest distance, z _max represents the furthest distance that can be detected, and grid_length represents the grid length Actual length.

The determined grid is such that each pixel represents an actual distance of 4cm, each grid represents an actual 12cm square grid, and the entire local map represents an environment with an actual range of 10m square. The result is shown in Figure 7.

Claims (7)

1. one kind builds method based on Kinect sensor depth map robot working environment uncertainty map, it is characterised in that: the method includes following steps:

Step (1): robot uses Kinect sensor sampling depth data；

Step (2): the depth data gathered is carried out pretreatment, obtains depth data figure；

Step (3): gather ground depth data and carry out ground model extraction；

Step (4): depth data figure is carried out ground model shear treatment and obtains barrier depth map, more former depth data figure is carried out shear treatment with barrier depth map obtain ground depth map；

Step (5): barrier depth map is carried out detection cognitive disorders region, ground depth map is detected and identifies clear area；

The uncertain grating map forming robot working environment is analyzed in barrier and free area by step (6)。

2. build method based on Kinect sensor depth map robot working environment uncertainty map according to claim 1: it is characterized in that: described step (3) is sampling depth figure under the environment of a spacious clear for the method adopted of extracting of ground model, image-forming principle according to Kinect, it is appreciated that depth image has the property that (1) is unrelated with the feature of image, only and distance dependent。(2) gray-value variation direction is consistent with the z-axis direction, direction, visual field captured by Kinect depth camera, and along with the increase gray value of distance can become big。So detecting that with the Kinect depth information apart from identical ground be identical。When setting Kinect sensor is constant with ground relative altitude and the angle of pitch, at sampling depth figure under the environment of a spacious clear, after distance exceedes certain threshold value, Kinect can't detect the ground data in front, so only taking the Kinect ground data that can detect, being all considered as invalid data elsewhere and being designated as 0；Due to the restriction of the performance of Kinect sensor own, its for terrestrial information nearby gather relatively good, poor at a distance, the data that more near place collects are more complete, and the data that collect at a distance are imperfect and error is relatively big, so also it to be processed；The every a line of depth image have recorded and the ground depth information under Kinect same distance, it is entered the depth information of the every a line of line scans record, remove the depth information of invalid data, remaining data are weighted on average obtained final data and are the ground depth information at this place；A ground model template is recorded and generated to the data handling every a line well；Thus obtain a ground model；Data in ground model are saved under program root。

3. according to claim 1 build method based on Kinect sensor depth map robot working environment uncertainty map: it is characterized in that: firstly the need of depth information being mapped to colouring information so that image shows in step (2), showing that Kinect can be detected by coverage through experiment test is within 10 meters, so being mapped between 0 to 255 by 0 to 10 meters, it is mapped to color by distance thus realizing the display of depth map；Obtain depth information color diagram。

4. according to claim 1 build method based on Kinect sensor depth map robot working environment uncertainty map, it is characterised in that: the region of described step (5) map is divided into free area, barrier and unknown area；Free area is set to detected ground region, barrier be set to detected by have the region of barrier, zone of ignorance is set to other regions except ground and barrier；Grid information is recorded with structure；Including the status indicator of grid, the confidence level of grid, the color of grid；Concrete operations comprise the steps:

(1) clear area detection algorithm: the depth data and the obtained ground depth data that are collected by Kinect compare, if ground depth data and the difference of depth data that collects are less than certain threshold value, preserve ground depth data, otherwise data are set to 0；Obtained is ground depth information, maps that under world coordinate system and records shared grid information；

(2) barrier zone detection algorithm: the depth data and the obtained ground depth data that are collected by Kinect compare, if ground depth data and the difference of depth data that collects are less than certain threshold value, data are set to 0, otherwise retain the depth data collected；Obtained is barrier depth information, is mapped under world coordinate system and records shared grid information after being carried out depth data column scan analysis。

5. according to claim 4 build method based on Kinect sensor depth map robot working environment uncertainty map, it is characterised in that:

Barrier zone detection algorithm:

Barrier region detection algorithm concrete implementation step is as follows:

(1) depth data collected by Kinect and obtained ground depth data compare, if ground depth data and the difference of depth data that collects are less than certain threshold value, data are set to 0, otherwise retain the depth data collected；

(2) the depth map scanning method obtained is carried out column scan, arrange for first: when sweeping to first non-zero numeral, record the seed points that this numeral is first barrier, when sweeping to second non-zero numeral, compare with first, if both differences are less than certain threshold value, both merge into a seed points, and taking both meansigma methods is new seed points；If both differences exceed certain threshold value, record the latter for new seed points；Until scanning through string；The obstacle information of every string is recorded, including the number of barrier, the distance of barrier, the number of the pixel that barrier comprises, barrier top coordinate, barrier bottom coordinate with structure；

(3) constantly repeat the every terms of information that step 2 obtains all different barrier of all row, different barriers is judged, remove the number all barriers less than certain threshold value of the pixel that barrier comprises；

(4) obtaining, according to step 3, the pixel position that abscissa is image, vertical coordinate is the coordinate system of actual range；

(5) barrier that the coordinate system obtained according to step 4 is again converted under actual range coordinate shows；Need that image coordinate is tied to camera coordinate system and arrive the conversion of world coordinate system again；Formula (1) is utilized to obtain the coordinate barrier data are converted to world coordinate system from image coordinate system；

$dx=|v-v_0|\frac{depth(u,v)}{f_x}$ Formula (1)

Dz=depth (u, v)

Wherein in above formula, dx represents pixel (u, v) relative centre position (u₀,v₀) offset distance in the X direction, dz represents the depth distance that this point is corresponding, f_xRepresent the focal length of X-direction for the inner parameter of video camera, be set to a definite value；

(6) which grid is disturbance in judgement thing data belong under world coordinate system, and records this grid

Clear area detection algorithm:

Clear area detection algorithm concrete implementation step is as follows:

(1) depth data and the obtained ground depth data that are collected by Kinect compare, if ground depth data and the difference of depth data that collects are less than certain threshold value, preserve ground depth data, otherwise data are set to 0；

(2) formula (1) is utilized to obtain the coordinate ground data is converted to world coordinate system from image coordinate system；

(3) judge which grid is ground data belong under world coordinate system, and record this grid。

6. according to claim 1 build method based on Kinect sensor depth map robot working environment uncertainty map, it is characterized in that: determine that model analysis draws the formula determining barrier confidence level by the characteristic of Kinect sensor sampling depth data and barrier confidence level, obtain uncertain grating map。

It is as follows that barrier confidence level confirms that algorithm implements step:

(1) owing to the Kinect measurement adjusted the distance exists error so needing grid is carried out Confidence Analysis；Depth data detected by Kinect is along with the increase of distance, and error is also with becoming big；And there is certain proportion such as formula (3) between the two:

$\mathrm{\&sigma;}z=\left(\frac{m}{fb}\right)z^2*\mathrm{\&sigma;}d$

In above formula, σ z represents the distance error for z place, f represents the focal length of depth camera, and b represents the length of base (distance of infrared emission end and receiving terminal), and m represents normalized parameter, z represents actual grade distance, and σ d represents the pixel distance of 1/2nd。

(2) obtain barrier confidence level and determine model；If detecting there is barrier on this grid, dropping on the probability on grid is (grid_length-2* σ z)²/grid_length²；Obtain the formula (4) of computation grid confidence level；

$p=\frac{f_{1*}E+f_2*R}{f_1+f_2}$

$E=\frac{\mathrm{\&sigma;}Max-\mathrm{\&sigma;}z}{\mathrm{\&sigma;}Max}$

$R=\frac{{(grid\_length-2\mathrm{\&sigma;}z)}^2}{grid\_{\mathrm{length}}^2}$

$\mathrm{\&sigma;}Max=\left(\frac{m}{fb}\right){z_{\max }}^2*\mathrm{\&sigma;}d$

In above formula, p represents the confidence level of grid, f₁,f₂Represent two kinds of ratios shared by data affecting confidence level。E represents the error impact on grid confidence level, and R represents the impact on grid confidence level of the grid length, and σ Max represents the maximum error of farthest, z_maxRepresenting the maximum distance that can be detected by, grid_length represents the physical length of grid。

7. according to claim 6 build method based on Kinect sensor depth map robot working environment uncertainty map, it is characterized in that: fixed grid is that each pixel represents actual range 4cm, each grid represents the grid that reality is 12cm square, the environment that actual range is 10m square of whole local cartographic representation。