An FPGA Implementation of Insect-Inspired Motion Detector for High-Speed Vision Systems Tianguang Zhang 1 , Haiyan Wu 1 , Alexander Borst 2 , Kolja K¨ uhnlenz 1 and Martin Buss 1 1 Institute of Automatic Control Engineering (LSR) Technische Universit¨ at M¨ unchen D-80290 M¨ unchen, Germany 2 Department of Systems and Computational Neurobiology Max Planck Institute of Neurobiology Am Klopferspitz 18, D-82152 Martinsried, Germany Email: {tg.zhang, kolja.kuehnlenz, m.buss}@ieee.org, haiyan.wu@tum.de, borst@neuro.mpg.de Abstract— In this paper, an array of biologically inspired elementary motion detectors (EMDs) is implemented on an FPGA (Field Programmable Gate Array) platform. The well- known Reichardt-type EMD, modeling the insect’s visual signal processing system, is very sensitive to motion direction and has low computational cost. A modified structure of EMD is used to detect local optical flow. Six templates of receptive fields, according to the fly’s vision system, are designed for simple ego- motion estimation. The results of several typical experiments demonstrate local detection of optical flow and simple motion estimation under specific backgrounds. The performance of the real-time implementation is sufficient to deal with a video frame rate of 350 fps at 256 x 256 pixels resolution. The execution of the motion detection algorithm and the resulting time delay is only 0.25 μ s. This hardware is suited for obstacle detection, motion estimation and UAV/MAV attitude control. I. INTRODUCTION Highly accurate real-time stabilization and navigation of humanoids and vehicles is a major research focus of robotics and automation. An important aspect is the use of high-speed visual servoing control loops running at framerates of several 100Hz controlling and stabilizing the motion of the system. This paper contributes an implementation of a high-speed motion estimation model. A fly’s panoramic vision system comprises at its front end several thousand photoreceptors feeding into a 2D array of motion detecting neurons which the animal uses for dynamic visuomotor pose and gaze stabilization and navigation in 6 degrees of freedom. The Reichardt detector [1] [2] [3] [4] [5] [6] is a well-known model which describes, at an algorithmic level, the process of local motion detection in the fly, leading from non-directional input to a direction selective output. In a structure of the fly brain called ’lobula plate’ large neurons are found which integrate these local motion signals and additionally form extensive connections amongst themselves [7][8]. These neurons have large receptive fields and respond best to particular flow-fields such as occurring during certain maneuvers of the fly in free flight [9] [10]. In engineering applications such as robotics, driver assistance systems or surveillance systems, a camera system is usually used as a sensor to gather information about the environment. Motion perception based on the fly’s vision system is com- putationally cheap and, thus, particularly suited for real-time applications. In addition to optimizing the motion estimation algorithm, it is also required to select a suitable hardware. In [11] a new EMD circuit implemented on micro-air vehicles (MAV) has been designed by using Field Programmable Analog Array (FPAA). Besides, several EMDs-based models were developed based on Very-Large-Scale-Integrated (VLSI) cir- cuits. In [12], a low-power VLSI chip was described which consists of a one-dimensional array of EMDs to perform motion computation. In [6], a biologically inspired VLSI system for measurement of self-motion was introduced. Later on, Harrison designed and tested a single-chip analog VLSI sensor that could detect imminent collisions [13]. Recently, FPGAs are favoured by engineers to implement EMDs. In [14], a real time algorithm for estimating motion vectors has been implemented. In [15], a FPGA implementation of a bio-inspired visual sensor is introduced. However, these implementations perform motion estimation with relatively low resolution and low frame rate. In this paper, an array of biologically inspired EMDs is implemented on an FPGA platform exploiting its major advantages: the execution of massive truly parallel computa- tions in only one processing cycle and the pipeline structure in data processing. Based on [16] two new simple templates of receptive fields for rotation detection are proposed to facilitate specialized motion detection and cover all types of movement. The performance of the implementation is sufficient to deal with video frame rates of 350 fps or above for a frame size of 256 x 256. The remainder of this paper is organized as follows: firstly, in Section II, the basic and elaborated EMD models are introduced. In Section III, we added two templates of recep- 2008 IEEE International Conference on Robotics and Automation Pasadena, CA, USA, May 19-23, 2008 978-1-4244-1647-9/08/$25.00 ©2008 IEEE. 335