Penetration Point Detection for Autonomous Trench Excavation Based on Binocular Vision

Abstract

To autonomously detect the penetration point in the working area of trench excavation, a feature detection method of penetration point based on binocular cameras was proposed. First, the homogeneous coordinate transformation is established, which can convert the 3D point cloud of the excavation area from the camera coordinate system to the excavator global base coordinate system. Then, the global gradient consistency function is designed to describe the geometric feature of the penetration point of a trench, and the position coordinates of the penetration point are detected. Finally, the test of the penetration point detection of the excavation area is conducted. Within the range of the excavation operation, the maximum position error of the penetration point detection is less than 80 mm, and the average detection error is 46.2 mm, which proves that this method can effectively detect the penetration point.