Research On Plant Characteristic Detection Technology Based On Kinect

The three-dimensional data detection of plant has a great significance to the cultivation and management of agricultural plant.At present,the real-time sensor equipment(such as laser sensor,machine vision system,and so on),which is highly praised due to its high precision,but this kind of equipment is expensive and is difficult to service the equipment.However,Kinect sensor has the advantages of small size,low cost and light weight,it has been widely used in the detection of agricultural as realtime sensor.In this study,Kinect as a real-time sensor can acquire the color image and depth distance image of the scene simultaneously,and then extract the RGB data and depth distance data of the plant in the scene.In this paper,firstly image acquisition platform with Kinect was set up to realize the real-time data acquisition of potted plant,and further study was taken to the acquired images and date.The main contents of the study and the main conclusions obtained in the followings:(1)The plant detection platform was build based on the Kinect sensor.Platform of static images acquisition is made up of a Kinect sensor image acquired system,Kinect data transmission line connected to PC,software by configuring a device interface control documents of MATLAB software for image data acquisition and storage.Platform of dynamic images acquisition is made up of sliding block moving slider to imitate sprayer machine straight line moving,including slip form system,slip control system and image acquisition system.Test under the condition of indoor fluorescent lamp,and the platform obtain images can meet the requirements of plant trees characteristics test.(2)The method of plant color image extraction based on the K-means and the Knearest neighbor algorithm was proposed,and the method of depth data restoration based on K-nearest neighbor algorithm was proposed.Design of experiment selected 30 groups of medium-sized plant Plant0 as static scene,the segmentation of RGB image and K-means clustering,compared with single segmentation method,the mean value of segmentation error reduced respectively 12.12 and 41.48 per cent,and obtained the higher accuracy.For depth image,match the target plant based on the corresponding color images,make the loss and error data restoration based on K-nearest neighbor algorithm,the repaired success rate as high as 97.15 per cent.For the detection of plant during dynamic processing,Design of experiment simulated car plant testing platform.The experiment take the way of Kinect image acquisition adaptive vehicle speed method,the dynamic experiment was taken in the range from 0.1m/s to 1.0m/s,and emphatically analyzes the color and depth images of 0.1m/s,0.5m/s and 1.0m/s speed,the quality of color image was decreased by the increase of vehicle speed,but the quality of depth image was not influenced by speed.And the plant detection method also can be used for dynamic images,it can also reduce the vehicle speed on the quality of the color images(3)The quantitative study of plant trees characteristics test,including two aspects accuracy analysis: plant depth numerical measurement and calculation of horizontal projection area.The test of sample carton testing surface and medium green potted plant canopy level projection area,the relative error of sample carton depth data was less than 0.24%,and the relative error of horizontal projection area was less than 7.6%,medium green potted plant canopy level projection area measured with high consistency.And demonstrated that in this study the use of Kinect to obtain images of the plant,plant feature extraction and three-dimensional data of detection is achievable and effective,can satisfy the requirements of the current agricultural applications.