1. Introduction Collision Prediction via the CNN Universal Machine
Collision Prediction via the CNN Universal Machine
V. Gál and T. Roska
Analogical & Neural Computing Systems Laboratory at the Computer andAutomation Research Institute of the Hungarian Academy of Sciences
H-1111 Budapest Kende u. 13-17, Hungary
ABSTRACT: In this paper we present an analogic CNN algorithm that estimates thetime to an impending collision between an approaching object and the observer.Calculation is based on a context insensitive method, which is well known in neurobiology,using only two specific cues of the expanding two-dimensional image of the looming object.
1. Introduction
Predicting dangerous or advantageous situations has similar importance for animals and for machines. One ofthese situations is when the motion of an object could end up with a useful (for a predator) or disadvantageous(for a prey) collision with the observer. A growing body of evidence makes neurobiologists suppose that forsome animals (and humans) low-level visual information alone is enough for a remarkable estimation of the timeleft till an impending collision.[1],[2] Fast calculation even in an unfamiliar environment may have a life-savingimpact. According to recent theories, calculation can be based upon only two optical variables of loomingobjects, ensuring the relative context-insensitivity of the process.[3],[4] Precise information about the size(diameter) of the two-dimensional projection of the object and the rate of its expansion are enough to predict theso called time-to-collision(TTC), provided that the motion is at a constant speed.
The type and simplicity of the method allow for the development of CNN algorithms [5] for the problem ofcollision prediction. Our analogical algorithm consists of 3 by 3 linear templates and has no special CNNarchitecture requirements, so it can be implemented on existing visual microprocessors.[6],[7]
2. Estimating the ‘Time-to-Collision’ (TTC): Optical Geometry
A method, which is supposed to be used by the neural system of some animals for TTC estimation could bederived from the following equation:
TTC=
dvedge
, (1)
where d is the diameter and vedge is the velocity of the edge of the expanding image (Fig. 1.). As the objectapproaches, the increase in the relative rate of expansion is higher than that in the diameter.
There is a law well known in optical geometry (the notations in the equations below are defined in Figure 1.):
s(tleft)f
and we can write:
=
Dd(tleft)
(2)
s(tleft)=vtleft
d
d(tleft)=vedge.up vedge.downdt
(3)(4)
A New Real-time Collision Prediction Model for ...on Machine Perception, Peking University, 100871, ...Introduction Robot soccer games are very popular ...
Collision prediction for polyhedra under screw motions The prediction of ...(bottom) Introduction The calculation of collisions amongst moving 3D objects ...
Introduction As the ?eld of arti?cial ... 83% of Machine Learning for Collision Selection ...prediction (Tsochantaridis et al. 2005) may ...
Introduction As the ?eld of arti?cial ... 83% of Machine Learning for Collision Selection ...prediction (Tsochantaridis et al. 2005) may ...
热门文档