Study On Monocular Based Obstacle Detection In Unknown Indoor Environment

Advanced visual robot is the tendency which the robot develops. Getting the environment information by vision to carry on self-guidance is the most common guidance way for advanced organisms. But it's difficult for the mobile robot to use the visual information effectively due to restriction of computing and cognitive mechanism of the robot. The monocular vision sensor becomes the current hot spot application in mobile robot for its rich information, good invisibilities, and low cost. This paper researches on the obstacle detection based on the monocular vision in the unknown indoor environment. We judge and measure the obstacle through the visual information.This paper goes forward by the world leading Pioneer3-AT mobile robot as research platform. Pioneer3-AT mobile robot has efficient hardware and software, especially rich environmental information sensation sensor and the software which favors the development expands. Based on vision system theory, we constructed Pioneer3-AT mobile robot's vision guidance system. On this basis, research carries on the obstacle detection algorithm in the unknown indoor environment.In unknown indoor environment, obstacles are complex and non-structurized. The mobile robot is unable to recognize each kind of obstacle like humans. Mobile robot confirms the obstacle by detecting the feature of edge. Mobile robot takes the intersection of obstacle and the ground as the recognizing target and makes the intersection as the non-obstacle region boundary. Mobile robot doesn't need to identify the particular obstacle in environment, but only need to identify the non-obstacle region, then could be possible to travel freely. Therefore the way adopted in this study to examine the indoor obstacle should be simple and effective.This paper detects edge of obstacles through HSI color space. We reduce the affection of the light in the edge detection using the characteristics of hue, saturation and intensity of the separation. We improve the measure method of color difference in the HSI color space to avoid detection of false edges. We detect H, S, I components edge by an improved Sobel operator. To improve the accuracy of edge detection, non-maximum suppression and Otsu method are used to determine the optimal edge threshold value. Finally, single-precision point sets of the intersection of obstacle and the ground are obtained by template scans. Experimental results show that the edge detection of obstacles based on HSI color space has less effect by light in unknown indoor environment. Besides, there is high detection accuracy.We can obtain the location and distance of obstacle points through vision measure. On the basis of the linear and nonlinear model analysis of camera, we calibrate the camera several times by 2-D plane calibration method, and get the parameter of the camera which required in the measurement algorithm. We present a geometric space constrained vision measurement method to measure obstacle. The visual measurement proposed in this paper is simple and low calculation. In addition, the experimental result shows that the measurement has high accuracy.