A smart buffer for tracking using motion data

Abstract

Responsive vision is vision responding to the environment. The characteristics of a responsive system are: active response in dynamic environment, real time computation, and using multiple modalities in a multi-purpose system. The Vision Engine[LBKL91b] is a genenal purpose for general vision tasks. Early vision processing, e.g., optical flow and stereo is implemented in near real-time using the Datacube, producing dense displacement fields at near video rates, which are then transferred to a Transputer subsystem, where data dependent processing occurs in parallel on subimages.W e use the Vision Engine for complex processing under real-time constraint, the differences between the processing rates in a robotic system require smart buffers, objects that can buffer data between pemeption, reasoning and action processes. Smart buffers oger a simple interface between asynchronous processing tasks and simplify the structure of multiprocessor vision systems. We describe a simple motion tracker that uses a smart buffer to mediate between early and middle vision processing. The smart buffeer permits the system to sense during action by letting the sensing component accumulate visual data in the course of action. *The authors may be contacted at little@cs.ubc.ca.

Real-time multitarget tracking by a cooperative distributed vision system

“…While the system architecture consisting of multiple parallel processes with a common shared memory looks similar to the "whiteboard architecture" [39] and the "smart buffer" [40], the critical difference rests in that each variable in the dynamic memory stores a discrete temporal sequence of values and is associated with the following temporal interpolation and prediction functions (Fig. 7).…”

Real-Time Quantized Optical Flow

“…In correlation-based ow such as in 8,13,14] the motion for the pixel at x,y] in one frame to a successive frame is de ned to be the determined motion of the patch P of pixels centered at x,y], out of (2 + 1) (2 + 1) possible displacements (where is an arbitrary parameter dependent on the maximum expected motion in the image). We determine the correct motion of the patch of pixels by simulating the motion of the patch for each possible displacement of x,y] and considering a match strength for each displacement.…”

Describing motion for recognition

“…Motion stabilization could be accomplished a tracking system such as [6] that pursues a moving ob- Figure 2 shows eight images from a sequence of 18 (from 08 to 25) for this subject. Only one field in a frame is used, because of the substantial difference between the two fields caused by motion.…”