Human following robot using kinect

This study presents a three wheeled human following robot with a Microsoft Kinect sensor as its only sensor for following a human target and avoiding collision. This study eliminates certain limitations for human following robots such as limited operating area and requiring the user to wear intrusive sensors. The system starts by making the human target do the lock-on process to tell the robot which user to follow. The robot follows a user by detecting the hip of the user and acquiring the depth data of the hip for steering and maintaining distance. Several tests are made to characterize the robot, these test are: straight path test, varying curve radius test, and collision avoidance test. The goal of the straight path test is to compare the path taken by the robot from the path taken by the user. Though the robot is successful in following the user, it was noticed that there is an average deviation of 4 to 8 centimeters from the path of the user. For collision avoidance, edge detection algorithm is used for detecting objects and steering away from the object. The collision avoidance test ensured that the robot was able to avoid objects in front of it. In the tests conducted the robot successfully avoided obstacles of at least one meter in height but makes, at most, a 180 centimeter of deviation from the path of the user. There is also a possibility that the robot will follow a different user after losing the user in its field of view during the target reacquisition process. Although the robot is able to follow a user in a straight path the robot tend to undercut the path of the user during curved path even with a safe turning radius of 1.5 meter, this was seen in the varying curve radius test. The undercut problem is a scenario where the robot collides with the reference point while turning. This is caused by the design of the robot having no sensors on its sides. Several solutions were investigated to solve the undercut problem which includes implementation of reducing the maintaining distance of human following algorithm, Finite Impulse Response (FIR) filter with weights designed by hamming window, moving average and weighted moving average and the simulation of the Target Following Algorithm using Arcs and Modified Target Following Algorithm. It was found that reducing the maintaining distance was able to make a difference of 50 centimeter in the entry of the robot to the path. The filters only smoothen the steering angle data but were not able to improve the undercut problem. Target Following Algorithm using Arcs can improve human following algorithm by 19.75%. while the Modified target following algorithm can improve the human following algorithm by 16.33%