AI for Autonomous Vehicles and Video Games3198285720000
Perception Module using Computer Vision. SLAM methods. Path Planning. Decision Making in First-Person Shooter
The state-of-the-art technologies are able to get humans away from the steering wheel and to take people into the gaming world. For these purposes are such technologies as computer vision, SLAM methods, path planning and decision making in First-Person Shooter are very useful.
Perception Module using Computer Vision
The automotive industry is considered to be a "pioneer" in machine vision field and its largest consumer. According to analysts, the automobile industry forms 23% of the market of products of computer sight in Germany. And according to VDMA, for Europe this figure makes 21%.
Using sensors and cameras, cars have learned to distinguish sides, trees, columns and the parked transport around themselves. The principle of determination of distance to objects is based on parallax – a method of determining the distance to an object using a simple formula:805692187545
In this formula L is for a distance between viewpoints, and ? is for an offset angle.
Another modern technique which is used in computer vision is the structure from motion (SfM). It involves estimating three-dimensional structures from two-dimensional image sequences, that may also be coupled with local motion signals.
Shortly, a stereo pair of images allows to identify pairs of corresponding points, and those allow to calculate the distance between the projected point in the three-dimensional world and the cameras.
A stereo pair recorded by two cameras / photo HYPERLINK "https://www.flickr.com/photos/[email protected]/15797745483/"https://www.flickr.com/photos/[email protected]/15797745483/
After many years of researching the science of stereo pairs, the automotive industry has all the tools for producing proper stereo images sequences. Calculated distance values can be visualised in a distance map. The distance to an object is painted by colours – the farther the object is, the more darkly is a distance map.3692136167572
Distance Maps calculated by applying a dynamic programming stereo algorithm / photo Darren Troy, University of Auckland
Simultaneous localization and mapping is a problem of composing and updating map of an unknown location while keeping track of an agent’s location within it. There is a wide range of SLAM Methods, while the most common solutions of this problem include particle filter, Kalman filter, and GraphSLAM.
Partical filters methods involve Monte Carlo method — a recursive algorithm for the numerical solution of estimating (filtration, smoothing) problems. Assume46559961854783845699197827
are Markov process and observations respectively.