The Monocular Vision-based Obstacle Detection For Autonomous Driving

Since the car has been created, both of the number of it and the technology about it havebeen development at a very high speed. The number of cars is rapid growth in recent years.While promoting technology development, it has also brought a lot of problems. With thepromotion of the convenience and safety, the self-driving technology is development rapidly.For the self-driving is based on obstacle detection, it is so significance to research on thistechnology.This article focuses on the obstacle detection in automotive autonomous driving. Beforestarting to research about obstacle detection, this paper did a lot of observation and analysisfirstly. After that this paper proposed several features of obstacles, which are pervasive andvery prominent features. Firstly, an obstacle is in the feasible region. Secondly obstaclefeasible region is against the background of great contrast. Finally the obstacle is generallyrelated with the shadow. These characteristics have a common prerequisite that theseobstacles are with respect to the autonomous driving. Only in this context, thesecharacteristics of obstacles are right and meaningful.When detecting obstacle, using the character that obstacle are in the feasible region, notonly can reduce the range of detection, then improve the detection efficiency. Meanwhile,within the feasible region, the other features used to get the obstacle in this paper can be moreeffectively. It means that after using the feature that the obstacles associated with the feasibleregion, the accuracy of the method proposed in this paper is improved. After getting the roughfeasible region, we believe that all obstacles that the car cornered when moving must be in therange of feasible region. At the same time, we also believe that the obstacle’s grain and colorare must difference the feasible region’s one. In this article this feature is expressed using thefeature points. With the above two features a lot of obstacles can be got in fact, but only usingthese features some interference just like signs of the road cannot be excluded. Therefore, thefeature shadow of the obstacle is introduction to correct the interference.Based characteristics proposed above, this paper proposes a method of detecting obstacle.This method first get the feasible region using a method proposed in this paper. It gets thefeasible region by growing the edge of the feasible region range. Then get the feature pointsof image using SUSAN method, and finally get then shadow range by a method raised by thispaper. At last we can finally get all the obstacles by using all the features above.In this paper, the characteristics which used to get obstacles are as universal as possible. When get these features, this paper use some basic and practical methods also. This paper tryto use an effective method to solve the problem, because the efficiency is a very importantthing in autonomous driving. After many tests, we find that using the features and method,which put forward in this paper, can effectively detect the obstacles in the image. Undernormal lighting conditions, this method is reliable and efficiently. This paper ignore theextreme weather, just like fog, night and so on.For using ordinary camera, the purpose of getobstacles is almost impossible to complete.