Dynamic Obstacle Detection Based On Panoramic Vision In The Moving State Of Agricultural Machineries

he manner in which the obstacles around the automatic navigation agricultural machinery can be detected and identified is of great importance in improving the safety and operation efficiency of agricultural machineries. This study was based on a panoramic camera to rapidly detect the dynamic obstacles around the moving agricultural machinery. It used the Lucas?Kanade optical flow algorithm to detect the moving obstacles in a panoramic image. On the basis of the actual farmland operation, background optical flow dynamic model was established to filter the background optical flow. The noise optical flow was filtered by combining K-means clustering segmentation algorithm and the variances of optical flow direction and of optical flow length within clusters. On the basis of the segmentation clusters, we used an external rectangular box to select foreground moving object. The principal optical flow direction in the segmentation clusters, and the distance between clusters determined whether the same foreground moving object was selected. Subsequently, we placed the corresponding combination processing into use, which can make the box select the complete foreground motion target. By processing 100 frames of the images, result showed that the average time consumption of the proposed method was 0.801 s. The accuracy rate of detecting dynamic agricultural machinery was 88.06%, the accuracy rate of detecting pedestrians was 81.61%, and the overall accuracy rate was 82.93%. Thus, the proposed method could meet the requirements of actual farmland operations and have a good instantaneity and detection results.