Human tracking by using multiple methods and weighted products

AbstractConventional wheelchairs are used by people who can actuate their hand, but there are conditions that cause the user cannot operate the wheelchair, such as handicapped people and amputated arms. Hence, they need an assistant to operate the wheelchair, as we know those conditions are dependent on others. An intelligent wheelchair becomes a solution to this problem by tracking and following an assistant in front of a wheelchair; it uses a camera that has been embedded there. So the assistant does not need to move it. In order to track a human, an algorithm that can detect a human or object is needed. Some algorithms with different approaches have been proposed such as SIFT, SURF, BRISK, AKAZE, KAZE, and ORB. Each algorithm has its own excess and feebleness. This research devises multiple methods for a human tracking for smart wheelchair, which aims to cover the method drawbacks with the advantages of other methods. The proposed method is performed using multiple methods mentioned above, which have sorted by investigating the method score. Subsequently, the first method is used to track a human and yield keypoints. If the keypoints detected are less than the threshold value, then the next method is used and so on until it reaches the threshold value. The threshold value is obtained from the average of keypoints detected in the first frame. This devised method has been evaluated by using 6 videos recorded by us and 10 videos from Visual Tracker Benchmark. This proposed method achieved an average accuracy up to 0.848609 on 6 videos and 0.663 on OTB dataset.