The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec)

This research is concerned with a fish tracking system. which employs the raw-image (unprocessed gray-scale image) of a scene. a global/local search of a genetic algorithm (GA) in combination with a models shaping the target of known shape. This GA is used in order to improve the real-time tracking performance of a manipulator directed toward the target of unpredictable movement. The raw-image is used since it does not increase the original noise. and moreover does not require any filtering processing time. The global search feature of the (GA) is used as a pre-processor for the search of a target together with a GA-based local search technique that focuses on interesting points find by the pre-processor in order to perform an intensive local search. The local GA search relies on a mutation technique of positional and orientational bit-strings in order to improve the GA-based scene recognition performance in terms of fast and reliable recognition of a target so as to increase the tracking performance. In order to evaluate the effectiveness of the proposed fish tracking strategy experiments have been performed using a 6-DOF manipulator and a fish in its natural scene. The results have shown the effectiveness of the proposed method.

View full abstract

Show abstractHide abstract

The tracking of multiple moving objects in image sequences is an important task for vehicle guidance, following moving objects, robot control and so on. In order to implement of these tracking systems, the moving characteristics such as position and velocity of moving object from image sequences must be considered. In most systems, the first step in tracking of object is to separate the foreground from the background or to detect it's motion. But current methods for separating the foreground from background are dependent on illumination (daylight, clouds, shadows) change, processing time and speed, direction or texture of object. In this paper, we deal with tracking of multiple moving objects by the foreground prediction and the background adaptation using Kalman Filter. The foreground prediction is necessary to processing time reduction by re-setting of tracking window and solving problem of mixture, split, mergence for multiple objects tracking and the background adaptation is an essential element for dectecting foreground of independently moving objects regardless of their speed, direction or texture, and so on. Also, based on Kalman Filter, an algorithm for optimal foreground detection and background adaptation is proposed, and it is composed of three modules of BCM (Block Checking Module), OMPM (Object Movement Prediction Module), and ABEM (Adaptive Background Estimation Module). The BCM detects a new object from image sequences and the OMPM predicts the position of moving object by Kalman Filter. Usually, the illumination change should be considered in the background estimation and it should not be detected as foreground and the last module, ABEM, plays an important role in that. The effectiveness of the proposed algorithm is proven through the experimental results for an experimental setup that two objects body move continuously and a CCD camera captures the objects with a fixed focal length. As results, the presented approach shows that it can well track the multiple objects and adapt the illumination change of background simultaneously.

View full abstract

Download PDF (407K)

Register with J-STAGE for free!