下载  >  人工智能  >  深度学习  > 深度学习在无人驾驶机器视觉上的应用

深度学习在无人驾驶机器视觉上的应用 评分

深度学习在无人驾驶领域主要用于图像处理,也就是摄像头上面。当然也可以用于雷达的数据处理,但是基于图像极大丰富的信息以及难以手工建模的特性,深度学习能最大限度的发挥其优势。
Images are numbers MENA 1949象9405781185760.871740943.6 6620 49317355791499371 309349133665 753。46011青 5637323号百 3216T 9419234224040263130 342447s353243642035171250 8812864236710263406759s4706616386470 2606826212309639363094096619923 2355956673996时7177796315悲34非96372 6397590764424351406:3343139s 12225316?1s946自16149535 16390542963531475558824001752436298557 656048357189070544376021851541758 1980160659447692873921865237770498540 s2335991607975732162626793327p6 ?62÷072346当?655136595969 044216 112497210“629 106g、E 641T250 。73352978319001743149T 2357955 771:556?192334根35:9 What the computer sees 829 cat image classification 15% dog 2%o hat 1% mu Regression: The output variable takes continuous values Classification The output variable takes class labels Underneath it may still produce continuous values such as probability of belonging to a particular class Massachusetts Institute of References:891 MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Technology https://selfdrivingcars.mitedu ex. mit. edu 2018 Computer Vision with Deep Learning Our intuition about what's"hardis flawed (in complicated ways Visual perception: 540,000,000 years of data Bipedal movement: 230,000,000 years of data Abstract thought: 100,000 years of data Prediction: Dog Distortion Prediction: Ostrich Encoded in the large, highly evolve sensory and motor portions of the human brain is a billion years of experience about the nature of the world and how to survive in it. Abstract thought, though, is a new trick, perhaps less than 100 thousand years old. We have not yet mastered it. It is not all that intrinsically difficult; it just seems so when we do it Hans Moravec, Mind Children (1988 Massachusetts Institute of References: [6, 7, 11, 68 MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Technology https://selfdrivingcars.mitedu ex. mit. edu 2018 Neuron: Biological Inspiration for Computation Impulses carried Differences(among others) toward cell bod branches Parameters: human brains have w10.000000 dendrites of axon times synapses than artificial neural networks axon nucleus axon Topology Human brains have no layers terminals Topology is complicated impulses carried away from cell body Async: The human brain works cell body asynchronously, anNs work synchronously Neuron: computational building Learning algorithm ANNs use gradient block for the brain descent for learning Human brains use.(we don ' t know) Processing speed: Single biological neurons are slow while standard neurons in anns are co U00 fast synapse axon trom a neuron 00 Power consumption: Biological neural dendrite cell body Wiii+b networks use very little power compared to 011 02x2+b artificial networks output axon activation Stages: Biological networks usually don't stop w2 / start learning ANNs have different fitting (train)and prediction (evaluate) phases (Artificial) Neuron: computational building block for the"neural network Similarity(among others) Distributed computation on a large scale Massachusetts Institute of For the ful updated ist of references visit: [18, 143] MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Technology https://selfdrivingcars.mitedu ex. mit. edu 2018 The Reticular Formation Radiations to cerebral Human vision cortex Its structure is instructive and inspiring Thalamocortical System Simulation: 8 million cortical neurons 2 billion synapses 】:D Visual impulses Auditory Reticular formation impulses Ascending general sensory tracts Descending (touch, pain, temperature) motor projections to spinal cord IIIr Massachusetts Institute of References: [118 MIT 6. S094: Deep Learning for Self-Driving Cars Lex Fridman January Technology https://selfdrivingcars.mitedu ex. mit. edu 2018 Visual cortex (Its structure is Instructive and Inspiring TOm Owes 5 Bill male ir : WJane TECH Sue cute rat V1 V2 V4 IT-posterior IT-anterior visual cortex Retina Layer 1 Layer 2 Layer L Pixels Edges Primitive Shapes Massachusetts MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Institute of Reference:https://www.youtube.com/watch?v=33k1zttoOw Technology https://selfdrivingcars.mitedu ex. mit. edu 2018 Deep Learning is Hard: Illumination Variability January Institute of MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman Technology https:/iseLfdrivingcars.mitedu/references[66 https://selfdrivingcars.mitedu ex. mit. edu 2018 Deep Learning is Hard: Pose variability Figure 1. The deformable and truncated cat. Cats exhibit(al Parkhi et al. The truth about cats and dogs 2011 Institute of MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Technology https:/iseLfdrivingcars.mitedu/references[69 https://selfdrivingcars.mitedu ex. mit. edu 2018 Deep Learning is Hard: Intra-Class variability Abyssinian Bengal Bombay I Persian Egyptian Fried ragdoll Eng. setter Boxer Keeshond Chihuahua Great Pyrenees German shorthaired Parkhi et al. Cats and dogs. 2012 Massachusetts Institute of For the ful updated ist of references visit: [70] MIT 6.S094: Deep Learning for Self-Driving Cars Lex fridman January Technology https://selfdrivingcars.mitedu ex. mit. edu 2018

...展开详情
所需积分/C币:16 上传时间:2018-08-29 资源大小:7.36MB
举报 举报 收藏 收藏 (2)
分享 分享