In the process of dismounting and assembling the drop switch for the high-voltage electric power live line working (EPL2W) robot, one of the key problems is the precision of positioning for manipulators, gripper and the bolts used to fix drop switch. To solve it, we study the binocular vision system theory of the robot and the characteristic of dismounting and assembling drop switch. We propose a coarse-to-fine image registration algorithm based on image correlation, which can improve the positioning precision of manipulators and bolt significantly. The algorithm performs the following three steps: firstly, the target points are marked respectively in the right and left visions, and then the system judges whether the target point in right vision can satisfy the lowest registration accuracy by using the similarity of target points’ backgrounds in right and left visions, this is a typical coarse-to-fine strategy; secondly, the system calculates the epipolar line, and then the regional sequence existing matching points is generated according to neighborhood of epipolar line, the optimal matching image is confirmed by calculating the similarity between template image in left vision and the region in regional sequence according to correlation matching; finally, the precise coordinates of target points in right and left visions are calculated according to the optimal matching image. The experiment results indicate that the positioning accuracy of image coordinate is within 2 pixels, the positioning accuracy in the world coordinate system is within 3 mm, the positioning accuracy of binocular vision satisfies the requirement dismounting and assembling the drop switch.
The most concerned problem is to detect the interesting objects in image sequence captured from the same scene. Image difference is a commonly used method in detecting the interesting object, however, massive noise exists in the binarized difference image, so how to remove the noise is a hot issue. Aiming at the removing the noise in binary difference image, we propose a novel filtering algorithm based on Haar feature density map. Firstly, calculate the Haar feature density distribution map of binary image. Secondly, the density distribution map of Haar feature is binarized to remove noise. Finally, the interesting objects can be easily detected. Experiments show that the Haar feature density map achieves a better filtering effect than the conventional filtering algorithms for binary image (such as median filtering, morphological operation and so on).