Obstacle Avoidance Strategy For Monocular Robot In Haze-fog Condition

The rapid development of artificial intelligence has promoted the popularization and application of robots,while robots can assist and replace human beings to work in some areas.For example,visual robots can replace human beings to accomplish dangerous rescue tasks in dangerous environment in disaster fighting and rescue.Visual robot obstacle avoidance is an important part of robot navigation.However,the images collected by the equipment was degraded in the haze environment,which lead to the weakening or even loss of the depth information and obstacle features used for obstacle avoidance in the images.Therefore,the research on obstacle avoidance strategy of monocular robot under the condition of haze was carried out.The key of obstacle avoidance is to obtain the depth information of obstacles.Relative to the method of getting the depth of obstacles under the condition of no fog scene,the depth information of the corresponding scene was obtained by the depth map of hazy image.But it was found that the fog haze degradation model of atmospheric optical value estimate was not accurate in the experiment,which meant that there were the errors between the actual scene depth map and calculation of depth map.At the same time,the matching precision was low and the cost was higher in the detection of obstacles of recognition.To solve above problems,this paper put forward the strong robust atmospheric optical estimation method at first,which mainly through the choice of image grayscale pixels and minimum filtering operation interference filter to estimate the haze atmospheric optical image value.After analyzing the error between estimated distance and precise distance in the fog conditions,the depth of the compensation model was proposed based on the characteristics of learning.By fusion edge information and brightness based on Poisson-editing fusion and training the image data in a data set to add fog treatment,accurate depth map was compensated by the specific parameters of the model.Finally,the camera imaging model of obstacle avoidance strategy was proposed for obstacles assumptions of the model.After getting the accurate depth map,the imaging model of camera in the scene was set up through calibration blind area location information and the tilt angle of the camera.Thus barrier-free areas were determined with the help of obstacles horizontal scan lines after on the depth of image edge detection and the path planning of next step was got through the key points of the model indicates that.In the experimental results,this method basically realized obstacle avoidance.The main contributions of this paper include:(1)The strong robust atmospheric light estimation method was proposed to estimate the atmospheric light value in the haze model,which improved the accuracy of atmospheric light value estimation;(2)To build error compensation model by analyzing the error between estimated distance and accurate distance under haze conditions;(3)By establishing the imaging model of the camera,blind area,optical center and other parameters in the scene,special features of the calibration feature points and the horizontal scan lines were used to achieve obstacle detection and path planning for obstacles under hazy conditions.