Research Progress of Intelligent Obstacle Detection Methods of Vehicles and Their Application on Agriculture
He Yong,Jiang Hao,Fang Hui,Wang Yu,Liu Yufei
DOI: https://doi.org/10.11975/j.issn.1002-6819.2018.09.003
2018-01-01
Abstract:The application of automatic navigation technology can improve the accuracy and safety of agricultural operation, and obstacle detection is an important part. According to different sensor measurement technologies, the detection methods of farmland obstacles in intelligent agricultural vehicles are reviewed, and the advantages and disadvantages of each method are analyzed. The single sensor measurement technologies include: 1) Ultrasonic measurement technology. This technology has the advantages of simple operation and low cost when used in obstacle detection; what's more, under certain conditions, it can detect obstacles in dark, dust, smoke, electromagnetic interference, toxic and other harsh environments. But it is easily affected by the surface condition of different kinds of obstacles, so this technology can be applied to the scene which includes only one kind of obstacle. 2) Laser radar measurement technology. Laser radar can be divided into 2 kinds: three-dimensional laser radar and two-dimensional laser radar. When used in obstacle detection, three-dimensional laser radar has the advantages of high accuracy and long detection distance, but has the disadvantage of high cost, while two-dimensional laser radar with low price has small perspective. Recently, researchers begin to convert two-dimensional laser radar into three-dimensional laser radar and have done some experiments in simple outdoor environment, but no experiment was done in farmland. 3) Machine vision measurement technology. This technology can get comprehensive image information when used in obstacle detection in farmland, but needs high computing power of computers and long detection distance of cameras. Recently, researchers mainly focus on the use of new image segmentation algorithm and stereo matching algorithm and many experiments were done in farmland obstacle detection. Meanwhile, aiming at the problem that single sensor measurement technologies cannot meet the needs of complex farmland environments, several kinds of multi-sensor fusion technologies applied in farmland obstacle detection are summarized, including the fusion of vision measurement technology and LIDAR (light detection and ranging) measurement technology, the fusion of vision measurement technology and ultrasonic measurement technology, and Kinect sensor measurement technology which combines depth and color image information. Finally, the existing technologies are analyzed and the future research is prospected, including the application of new method (equipment) and the improvements of existing methods. The application of new methods can be concluded as follows: 1) ZED stereo camera (detection distance ranges from 0.4 to 20 m) and low cost laser radar. 2) Use new algorithm (like deep learning) in obstacle detection and recognition. 3) Agricultural unmanned aerial vehicles. And the improvements include: 1) Improve obstacle detection algorithm to find the balance between accuracy and processing time. 2) Use some technologies, like hyperspectral image technology, to classify different kinds of obstacles before detecting.