Have you ever wondered how many hours does a person spend driving in his entire life?
Don’t worry. I got you covered.
According to a survey, In America, if a person starts driving at the age of 17 and drives until the person is nearly 79 then the person spends a whopping 37,000 hours driving the car throughout his/her life covering 1,284,256 kilometers, approximately it is the distance taken to drive to the moon and back — three times.
Imagine… if not completely, at least a part of the 37k hours being saved. Not by taking other transport mode or by appointing a driver but by opting for a self-driving car. YES!
These self-driving cars are capable of driving without any human intervention. For few, It might sound a little spooky but it’s not. Thanks to Artificial Intelligence. These cars are built to use a large amount of data from image recognition systems, neural networks, along with machine learning to drive autonomously.
These cars not only save time but they do save lives too.
According to another study, around the globe, nearly 1.3 million people die of road accidents each year i.,e.. on average 3,287 deaths per day.
94% of such accidents were caused by human error, 2% caused by environment, 2% by the vehicles and 2% by unknown factors.
And such 94% of human error can be avoided by these cars as these cars do not get distracted, they obey traffic rules, do not drive fast, do not drive drunk, they do not fall asleep (DUH!)
· These self-driving cars are capable of adjusting the vehicle’s speed based on the speed of others vehicles and objects on the road, using automatic brake systems in times of emergencies making them safe for use.
· automatic parking systems, autopilot systems making them user-friendly
· its capacity to read road-signs making it adaptable and many other features which make these cars more safe to not just the person driving but for others as well.
But how does the car manage to identify the road signs, other vehicles on the road and people too?
The LIDARs and other sensors in the car, using the computer vision, translate an image into a relevant object, features or patterns for the processing unit of the car where the driving decisions are taken. This is where Deep learning comes into the picture. Deep Learning is a machine learning technique, learning features directly from data and the methods of deep learning are proven to be more accurate than human in classifying images. So the more and more the machine works on the image the more and more features of the objects are learnt by the machine.
A machine learning model involving the training data is built by a developer for training the machine. To make this building process quicker and easy a deep learning framework is used. This framework provides components for designing, training and validating deep neural networks which use mathematical modelling to process data. Some of the popular DL frameworks are Tensorflow and keras(google), mxnet (Amazon), PyTorch (Facebook), Caffe ( Berkeley).
To make it simpler, With the help of an image recognition system, large amounts of data in the form of images are fed to the machine(car) to recognise the objects and people.
But before feeding to the machine, these images go through a process called annotation.
Annotation is a process of labelling and annotating the data for training the computer models to recognize the objects on the road. This process gives an accurate reference to coordinate with the reality of the car’s surroundings. Such annotation is done using the [G15] rectangle tool, cuboidal tool, polygon tool, circle, point and bound boxing tool. These tools are selected based on the shape of the object. [G16]
Vehicles like cars and trucks can be annotated using the cuboidal tool. This gives a 3D effect to a 2D image.
As these road signs are normally square or rectangle we can use the rectangle tool
Polygon tool can be used to annotate human as they do not have a definite shape.
In this process, first the object is labelled with a name and then one of these tools are used for annotating the object in the image.
To make annotation walkover cake we have OCLAVI. It provides
Tools to annotate objects even with low-quality images with precision.
It gives the developer a choice of approaching the company’s freelancer or the developer can even bring his own team.
It allows the developer to connect to cloud storage for getting things done without any interruption in model building.
These self-driving cars can be called as a revolution in the industry of automobiles and thanks to AI for such a revolutionary technology. Such revolutions are a boon to the mankind and it is our part to use such technology to the fullest making the right use of it.
Explore more OCLAVI
Source: Deep Learning on Medium