Personal ROADAR

Published on June 06, 2014

Personal ROADAR

Marat Bashlykov, co-author and ideological inspirer of the RoadAR project, relied on his own need to notice signs on the road. Traveling a lot by car, he noticed that the driver does not always see signs, and this threatens with fines and generally increases the risk of driving. On the other hand, many have a video recorder whose camera is capable of capturing signs, and a phone that can process data from the camera. So the idea came up to create RoadAR.Last year, the project participated in the acceleration program GenerationS, went to the finals of the Web Ready contest, and this year won the Intel Startup Village conference and intends to represent Russia at the Intel Global Challenge in California.

S.K .: Marat, tell us how RoadAR works?

M.B .: RoadAR (note: a phone equipped with the RoadAR app is placed on the holder in the car, starts and starts recording video, while analyzing the image with computer vision algorithms. On a modern smartphone, with a frequency of 30 frames per second, it detects and recognizes traffic signs, understands their coverage area and warns the driver in case of a violation or a possible danger. There are many difficulties here: getting a smartphone to record video in parallel and analyzing it is no longer easy, but we still need to impose coordinates on the video (i.e. mix it), make recognition, and even so that it works on a huge fleet of android devices. The biggest problem now is energy consumption, but soon we expect to significantly reduce it by optimizing the operation of recognition algorithms.

S.K .: Do you use any third-party developments?

MB: We use OpenCV for some basic transformations and operations, but we write a lot ourselves to make it work fast enough on mobile devices. A lot of work has been done in the field of GPU computing, but so far we are not connecting it due to the same power consumption.

S.K .: With what accuracy does the system work? What affects the quality of recognition?

MB: There are 2 key points in character recognition (assuming that the smartphone is quite powerful and has a good camera) - the detection of a candidate for a character and, in fact, the recognition itself. Naturally, in the case of a smartphone, we cannot run through the window throughout the frame and look for signs there. Now we are looking for suitable forms and, when they are found, we give them to the “recognizer”. We have achieved a quick search for basic forms, but the only limitation now is that they must be closed. Those. overlapping often makes it difficult to find the shape, but the shadow does not. On the GPU, it turns out to look for any form and quickly enough, but power consumption is growing even stronger. Now we are working closely on a fast algorithm for finding forms on the CPU, which would not be so much affected by overlap.

The accuracy of recognition of candidates for most characters is very high, for some it is 99.7%, for some 95%, but, of course, there are difficulties. For example, the sign "give way" - it is often confused with wires and poles of power lines. The fact is that this is the only sign that is an inverted triangle and it does not have any special signs by which it could be distinguished and for which the algorithm would catch. We have many ideas on how to solve this problem, now we are engaged in testing them.

S.K .: How is the monetization model built?

MB: We do all the basic functionality for free. Now we are thinking about creating a few additional chips for money or for share, for example, a warning about traffic lights and pits. We can also collect a lot of information that is interesting for different businesses. Information from billboards, prices at gas stations, etc. We also collect specialized signs for freight transport, these are such signs as “weight restriction”, “height restriction”, “weight restriction per vehicle axle”, etc., “truck traffic is prohibited." All this information is necessary to draw up the optimal route, and logistics companies were already interested in such data with us. You can collect a large amount of various information - about traffic jams, accidents, quality of road infrastructure.

Another idea is advertising. Sound advertising that should be interesting to the driver. For example, in the morning and in the evening you stop at a kindergarten, therefore, you have a child and you may be interested in advertising children's products, respectively, when you approach the children's world, you will hear an offer from the store, which also contains a discount. Advertising can be useful and it should not be much.

S.K .: Tell us a little about the project team?

MB: We have a small team, 5 people. Eugene and I drive a lot, according to our previous work, we often went on business trips, and now we often go to competitions in other cities, and always by car. Faced with the fact that modern gadgets do not fully use their capabilities. We constantly thought and discussed how to improve them. We decided that step by step we can do a lot ourselves, and we can collect data for mapping using crowdsourcing. They quit their jobs, put together a team and sat down to work. Eugene called to us his former classmate Andrei, who at that time already had his own successful business, he worked for outsourcing with the USA and Australia, and made highly loaded decisions. He was also interested in our idea, so much so that he left his business to partners and came to our team.

Two very cool guys - Foat and Sasha. Foat is a specialist in computer vision and machine learning, we have been looking for him for a long time. We went to universities, talked with professors, they advised us of their students, but someone was afraid to leave their familiar place, someone did not suit us for other reasons. And when we met Foat, he was inspired by the idea, and he himself is not a driver, he was just interested to do what we are doing.

And we met Sasha by chance, so to speak, a combination of circumstances: a job fair was held in Kazan IT Park; we didn’t really hope for it, but closer to the end a young guy came up to us, said that he did not know what could be useful to us, but he was interested in sawing such a thing. So it works with us now, without him we are without hands :)
At the very beginning, everyone told us that what we had planned was impossible to do, smartphones could not cope with recognition. This was a challenge for us, and now we periodically prove that nothing is impossible, especially when such a well-coordinated team does it!

S.K .: How did you manage to find investments?

MB: We have big problems with investments. We do not do an online store where everything is clear: sales funnels, landings, CPA, LTV.
We have a complex monetization model and from the technological side everything is very clever. Apparently, many investors find risks in this. For six months we did everything with our own money, invested a fairly large amount of personal funds. Then three angels and the MSU Business Incubator invested in microinvestments. This money was enough to issue an official release and understand that people are interested in what we do. Over the past week, we have received 15K installations. Now we are looking for the next round of investment.

S.K .: Tell us about your impressions of participating in GenerationS?

MB: The most wonderful impressions are many good projects, interesting speakers, useful lectures. As a result, this is an excellent networking and the opportunity to chat with representatives of large companies such as Intel, Yandex, listen to David Yan and Natalia Kaspersky.

S.K .: “Where to run” to a young startup who has an idea but does not have the money / qualifications to implement it independently?

M.B .: To the wall. Wherever you go, get ready for a wall in front of you. But, if you believe in your idea - do not give up :)
We started by coming to the Business Incubator of Kazan IT Park, and this helped us a lot. Now there are many interesting acceleration programs, incubators, etc. There you can test your idea, find like-minded people, and, perhaps, even attract small investments for launch.



Interview for GenerationS contest was taken by Sergey Kokarev