New Turn: How Computer Vision Will Change Auto Industry
According to analysts’ estimate, first fully ‘self-driving’ car that will be able to take its owner from point A to point B will appear on the market by 2019. By 2020 global roads will have up to 10 million self-driving cars.
On of the most important ‘skills’ of an autopilot is ability to analyze situation on road and make a decision to act based on information obtained from camera. Special computer vision systems are use so the car is able to recognize objects it encounters on the road.
What Is It
Computer vision technologies have been developed during the last several decades. Scientists want to teach a computer to see and analyze information it gets.
Computer vision has a variety of applications. Such technologies can be used, for example, to get 3D models from photos and videos in the process of digital map creation, to create building architecture models or in security systems for facial recognition. But computer vision will have the highest impact on everyday life of people in automotive industry.
How Cars Can See
One of the most obvious applications for computer vision in vehicles is creation of accident prevention systems. Currently, a technology called (Mobileye)is actively developed and used in cars of BMW, General Motors and since recently in Tesla Motors. Swedish car maker Volvo is also developing their own technology called City Safety.
This system monitors car movement trajectory and warn the driver if he is too close to the car in front or in case of hard braking. That being said, on top of a warning there can be an automatic braking in critical situations operated by autonomous emergency breaking or AEB systems used in self-driving cars:
According to published in the end of January 2016 statistics from Insurance Institute for Highway Safety, IIHS, the use of automatic emergency braking allows to reduce the number of collisions with vehicles driving in front by 39%, and total number of accidents by 12%. In addition, Audi refers to data, according to which application of such systems allowed to reduce passenger injuries by 38% within five years.
There are other application for computer vision. For example, Ford has taught vehicle’s headlights to react to information obtained from frontal camera. Electronics spread the beams of headlights upon approaching intersections or roundabouts so that the driver can see objects moving along the sideway. In addition, system is able to identify suddenly appearing objects on the road by cast a light on them. General Motors is confident that this allows to reduce the probability of collision with an obstacle:
Computer vision technology is also used in creating navigation systems. With the help of cameras and sensors a vehicle can identify its location, refer to digital map and create a route. Such project, for instance, is developed by engineers from Mercedes-Benz (Route Pilot system).
Computer vision can be used not only to improve existing cars but in designing new ones too. For example, building 3D models of a car being designed from photos and videos.
In their turn insurance companies can use modern tools to record car damage from accident. Moreover, accidents involving cars equipped with computer vision can be reconstructed more easily.
Problems and Prospects
Despite all the benefits computer vision can bring and convincing statistics telling that its application will allow to increase safety on roads, there are certain obstacles to its advancement.
First of all because of a high complexity of necessary mathematic calculation: data from cameras and sensors are color values of 2D pixels per se based on which an algorithm has to build a 3D geometry of the scene, recognize and classify moving and stationary objects. System has to be able to recognize people and animals from stationery objects, consider peculiarities of lighting and reflection on mirror surfaces and do it all on the go.
Cameras are not always able to get a high quality image, which complicates processing. On top of that there are certain limitations in computing capabilities of available hardware as car manufacturers tend to avoid car price hikes because of new technologies.
Anyway, one cannot say that the existing problems are unsolvable, and many companies continue to work on technologies that will yield high precision of processing images from car camera without involving huge computation capacity. One of such technologies is being created by Cognitive Technologies together with KAMAZ in creating a self-driving truck using a principle of foveal vision. It allows a car to build ‘interest zones’ that shape a virtual tunnel - in this case the system can analyze only 5-7% of video without losing quality of analysis.
Unlike in the West, there are not that many companies engaged in computer vision in Russia, however, this direction is being developed. For example, on top of developing telematics solutions here at Smartdriving we are also developing computer vision technology.
Computer vision is not our primary business direction. Nonetheless, this technology is very promising and that is why we invest in its exploration and communicate with overseas researchers. For example, last year we visited International Conference on 3D Vision 2016 that took place in France and agreed to establish cooperation with specialists from one of the American institutes of technology.
The development of computer vision technologies and statistics collected by developers and third party organizations allows us to argue that within the next 5-10 years these techs will be able to substantially increase the road traffic safety. Even before self-driving cars will dominate the roads, smart braking, lighting and mapping systems will allow to dramatically reduce the number of accidents and casualties.
New automotive technologies such as telematics and computer vision will not only help increase road traffic safety but also open new development opportunities for related businesses, like insurance business. In a few years automotive market and related industries will experience dramatic changes. And it means that those who invest in the development of new solutions now will be able to become leaders in the era of new technologies.