Trash detection and tracking for a river cleaning robot

Abstract

The objective of this research work is to develop an algorithm for a river cleaning robot. This algorithm will be able to autonomously detect and track the surface trash flowing in water from a real time video. The state of the art works that have been implemented so far consider that the first frame of the video contains the object to be detected. The various challenges confronting this work are the reflecting nature of flowing objects in water, clutter and noise generated by ripples in water, changes in the HSL levels of the surface of the water and detecting multiple objects in a single frame. Our aim was to develop an algorithm that detects the trash flowing in the water, then keep a track of it as it progresses and then localize the object such that it can be collected by the scooping arm of the robot. This problem has been divided in two parts, where the first part concerns itself with the detection of moving trash and another with tracking the detected trash objects. We did a comparative study of two main detection algorithms, where one is the Background Subtraction in which the mobile object is detected by differencing the current frame from the reference frame of the video feed. The other detection technique is based on the Viola-Jones detection where the algorithm was trained to detect the surface trash. The Viola Jones algorithm could detect only one object in a given frame. Then the tracking was performed by using the kalman filtering and the KLT feature tracker. The feature tracker was unable to track sudden changes when compared to the Kalman Filter tracking. Thus the best features of the detection-tracking have been combined in our proposed algorithm. In this work we apply robust morphological processing with Background Subtraction to detect the active trash and kalman filter to track it. The algorithm is able to detect the object when it first appears in the frame, no training models are required as in the case of Viola-Jones method. Then the tracker is able to faithfully detect The object without losing it against the sudden currents of water.