Detection of Free Obstacle Region using Distance Transform and Image Subtraction Method for Monocular Camera Sensor

Abstract

An area that is free of any barriers is known as a free region. Prior studies have demonstrated that when textureless impediments are used, the pixel volume expansion approach has limitations in identifying free spaces. To address the shortcoming, an approach that combines distance transform and image subtraction techniques is suggested. This algorithm incorporates three primary image processing processes: distance transform, image subtraction, and k-means segmentation. An object that is 170 cm and another that is 200 cm from the camera serve as the algorithm's inputs. Following image processing, the output is separated into 12 sections, one of which is designated as a free region. Two distinct scenarios—a congested atmosphere and an uncluttered environment—were tested by the program. Four different types of obstacles are tested for each scenario. In this case, there are four types of obstacles: texture, textureless, textureless with one obstacle on the left, and textureless with one obstacle on the right. The outcome demonstrates that the technique can accurately identify free zones. Most importantly, it has a 100% success rate in detecting obstructions that lack texture. This demonstrates that the approach can get around the prior method's drawback. Additionally, each obstacle's success rate is compiled and shown in a table. Furthermore, a comparison is made between the proposed method and the pixel volume expansion method. Combining the image subtraction and distance transform method with the volume pixel expansion method yields the same success rate (100%) in texturing barriers for crowded situations.