Computer vision detection of negative obstacles with the Microsoft Kinect Wang, Luke; Vanderhout, Russell; Shi, Tim
The objectives of this project were to use the Microsoft Kinect to develop an obstacle detection system for wheelchair users and output the obstacle data in a way useful for future projects using the system. This project was done as part of the CanWheel project, which is meant to improve the mobility of old adults using wheelchairs. The Kinect is essentially a 3-D camera which can produce depth images, images in which each pixel contains distance data. The Kinect is attached to the back of the wheelchair and aimed towards the floor. Obstacles may then be found by using the depth data and locating parts of the depth image which do not belong to the floor. This system is designed for indoor use, as it requires that the ground is a flat surface. Also, since the depth camera transmits and receives infrared waves, sunlight causes a disturbance. The purpose of this report is to provide technical details related to the obstacle detection system, explain the method used, describe the results, and make future recommendations. Background information includes details of pinhole camera analysis and the stereo camera parallax. The algorithm used to find obstacles consists of detecting planes to find the floor, then determining which pixels correspond to obstacles. Finally, all the detected information is all mapped onto an overhead view. It is recommended that further optimization of computational resources is done with the application of multi-threading CPU, General Purpose GPU and FPGA. One may also modify the algorithm based on their specific use case to further optimize the software.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International