Visual Tree Detection for Autonomous Navigation in Forest Environment Wajid Ali, Fredrik Georgsson and Thomas Hellstr¨ om Abstract— This paper describes a classification based tree detection method for autonomous navigation of forest vehicles in forest environment. Fusion of color, and texture cues has been used to segment the image into tree trunk and back- ground objects. The segmentation of images into tree trunk and background objects is a challenging task due to high variations of illumination, effect of different color shades, non-homogeneous bark texture, shadows and foreshortening. To accomplish this, the approach has been to find the best combinations of color, and texture descriptors, and classification techniques. An additional task has been to estimate the distance between forest vehicle and the base of segmented trees using monocular vision. A simple heuristic distance measurement method is proposed that is based on pixel height and a reference width. The performance of various color and texture operators, and accuracy of classifiers has been evaluated using cross validation techniques. I. I NTRODUCTION Autonomous navigation of vehicles in off-road environ- ment has already got a great deal of attention in last few decades. One research project, the IFOR Navigation (Au- tonomous Navigation for Forest Machines) [1] is going on at the Department of Computing Science, Ume˚ a University. One important part involves detection and avoidance of known or new obstacles appearing on the path and on both sides of vehicle (forwarder shown in Fig. 1). In some cases e.g. if a tree on the left and the right side of vehicle is too close, the system should stop the vehicle and alert the remote human operator, who should be given the option of manually correcting the vehicle position, or giving the system the green signal to go ahead along the original recorded path. This paper presents a computer vision system that detects tree obstacles on the left and the right side of an autonomous forest vehicle and estimates the distance between a forest vehicle and the base of detected trees in order to make the navigation safe. The article is organized as follows: Section II describes the related work. Section III presents tree detection and distance estimation algorithm. Section IV provides the implementa- tion details. In section V, we present the experimental results followed by discussion. Section VI has the conclusions and future works. II. RELATED WORK Extensive research work has already been done in range- based obstacle detection, and there is also much work has been done in range and reflectance based hybrid obstacle W. Ali, F. Georgsson and T. Hellstr¨ om are with the Department of Computing Science, University of Ume˚ a, S-901 87 Ume˚ a, Sweden {int04wai,fredrikg,thomash}@cs.umu.se Vid e o Machine Vision Camera Forest Machine (Valmet 830) Field of view (FOV) Fig. 1. Valmet 830 Forwarder, a forest vehicle detection but considerable less attention is paid in only reflectance based object detection. Interestingly, the first au- tonomous mobile robot Shakey [15] had a simple appearance based obstacle detection in textureless background. Ulrich and Nourbaksh [4] have also developed appearance based obstacle detection system that is based on passive color vision. Another similar system has been created by Batavia and Sanjiv. They used only color information for object detection [2]. Korah has used vision based texture cues for desert road following [11]. Another interesting system developed by Ollis and Stentz [8] navigated an industrial alfalfa harvester by following the cut/uncut crop line in a field. To do this, a color image was taken of a field, and a best fit step function was computed for each image scan line using the Fisher linear discriminant in RGB space. More information on appearance based related work can be found in [4]. A range-based obstacle detection system has been developed by Bostelman, Hong and Madhavan, they have used only TOF range camera for object detection towards AGV safety and advanced navigation [14]. Many researchers have also fabricated obstacle detection systems based on hybrid appearance and range information. In this category, a system developed by Christopher [12] can be used. He has used laser range, color and texture cues for autonomous road following. Similarly, Manduchi, Castano, Talukder and Matthies have successfully used hybrid techniques in obsta- cle detection and classification for autonomous navigation in an off-road environment [10]. Another related algorithm, though the purpose was somewhat else has been created in the same context as the one we have used in our project.