Kinect Depth Image Based Door Detection for Autonomous Indoor Navigation Yimin Zhou 1 , Guolai Jiang 2 , Guoqing Xu 2 , Xinyu Wu 2 and Ludovic Krundel 3 Abstract—In this paper, an indoor navigation algorithm is proposed for the purpose of robot autonomous path planning. Due to the complex situation in indoor environments, it can cause a serious trouble for robot to identify the route during patrolling, especially for corner and door detection, which is the key step for intelligent navigation. To solve this problem, a kinect sensor is used for the door detection and corner location via depth images. The continuously varied ratios and depth difference in the images have been analyzed for the corner and door identification. Furthermore, the precise position of the doors and corners can be localized via the 3-dimensional characteristics of the depth images. Experiments in different scenarios have been performed to verify the efficacy of the algorithm for robot indoor autonomous navigation. I. INTRODUCTION Real-time obstacle avoidance methods and navigation techniques are the core technologies for mobile robots and intelligent robots [8]. Due to various applied environments, unpredictabilities and incomplete perception measures, au- tonomous robot navigation has been seriously constrained in practical applications. As for intelligent service provided by residential robots, how to successfully avoid the obstacle in all kinds of complex indoor environments, is a prerequisite for the accomplish- ment of tasks, i.e., cleaning and monitoring etc. It mainly relies on the door detection during the indoor navigation [15], which plays an important role in path identification. Particularly, it will decide whether the robot should enter the room via the status of the door, i.e., open or closed [10]. A probabilistic approach for door detection is adopted in [7], which can capture both the shape and appearance of the door. At present, non-visual distance sensors, i.e., Microsoft Kinect, Asus xtion etc are gradually adopted for the door detection. The robot could detect the door at the front through swirling itself so that the open door can be located. However, if the door is closed and the robot is not right in front of the door in an unknown environment, it would cause faulty diagnosis. Therefore, the door status detection and location is the premise for robot autonomous path searching and blind indoor navigation [5]. *This work is partially supported under the Shenzhen Science and Technology Innovation Commission Project Grant Ref. JCYJ20120615125931560. 1 Y.Zhou is with Shenzhen Institutes of Advanced Technology, Chinese A- cademy of Sciences, Shenzhen 518055, China and also with the Chinese U- niversity of Hong Kong, Hong Kong, China. ym.zhou@siat.ac.cn 2 G.Jiang, G.Xu and X.Wu are with Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, China, 518055. 3 L.Krundel is with School of Electronic, Electrical and Systems Engineering, Loughborough University, Leicestershire, LE11 3TU, UK. l.a.krundel@lboro.ac.uk The interior door detection methods can be classified into traditionally non-visual and visual-based methods. Ultrasonic information is adopted in [2], and the combined sonar and video is used in [1][13] for door detection. However, the sound waves are easily affected by the door reflection coefficient, and it will have negative effect on the final result accordingly. The edge detection with laser distance meter can be used for fast door detection [4], but a high-precision motor is the requirement which results in high cost. A method using context-based object recognition to detect doors is proposed [12], which tags the door shape and height of the doorknob as the recognition label. Since there are so many types of doorknobs, it could increase the difficulties for detection. In [9], the Hough Transform is adopted to extract the edge lines from the images. Based on the edge lines from the left-side, right-side and bottom of the door, the door can be determined via the fuzzy recognition system. However, window and bulletin board such objectives can be misdiagnosed as doors due to similar edge numbers in the fuzzy system. A door detection algorithm is proposed in [3] based on the concavity and bottom-edge intensity of the door, and it could result in misdiagnosis as well. Based on the low-order video image processing tech- niques, a hypothesis and verification feedback mechanism is used for corridor and door detection [11]. If there are lots of corners in a particular indoor environment, this method can still provide accurate detection. In [14], a door is modelled based on the door edge and corner characteristics so as to achieve the door detection. However, this method is limited in many situations due to different shapes and types of doors. A door modelling procedure is described and its accuracy is evaluated statistically [9][12]. During the modelling, the shape and color are used as parameters, which are sensitive to the varied lights. Detection could be failed only via color information due to too many similar colors of the objects. Combined with the color and depth images from Kinect sensor, the obtained images are analyzed for localization. In this paper, Kinect sensor is used for door and corner detection/location with only black and white depth images. The remainder of the paper is organized as follows. Section II introduces the principle and algorithm of detecting interior doors and corners. Section III discusses the experi- ments and results for indoor robot navigation. Summary and future works are given in Section IV. II. THE PROCEDURE OF INDOOR NAVIGATION Considering the inside layout of rooms and constraints of camera angle, it is seldom to view or detect two sides The 23rd IEEE International Symposium on Robot and Human Interactive Communication August 25-29, 2014. Edinburgh, Scotland, UK, 978-1-4799-6765-0/14/$31.00 ©2014 IEEE 147