A product idea for a startup – A Traffic Sign Recognizer

Product ideas that are waiting for a scientific advance are not ideal for most startups! You cannot get such advances merely by putting pressure on the staff! Nor are ones requiring a whole lot of technical R & D. In contrast, there are ideas tried and tested as reported by a set of papers.  One such is a traffic sign detector and recognizer. Can you visualize a car mounted video camera that looks out for traffic signs such as stop signs, speed limit displays, “don’t enter” signs, children crossing warnings? A number of researchers have shown this can be done. Visit

Traffic Sign Detection and Pattern Recognition Using Support Vector Machine
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4782748&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D4782748 and

Road and Traffic Sign Detection and Recognition
http://www.iasi.cnr.it/ewgt/16conference/ID31.pdf

One announcement on the Web claims that the reported solution performs at superhuman level.

 First Superhuman Pattern Recognition
http://www.idsia.ch/~juergen/superhumanpatternrecognition.html

Use of colour for detecting and segmenting the traffic sign image from the overall scene is quite popular. Use of Support Vector Machines is a popular technique and so is the use of neural nets. It is clearly within reach to detect and recognize traffic signs from good quality photographs, and to give lab level demonstrations of the techniques.

Use of visual pattern recognition offers the big advantage that a product does not have to wait for the world to be re-organized. Traffic signs that are in wide use for human recognition serve as the environment for camera based techniques. Imagine asking a city to add even one LED to each traffic sign and to ensure regular power supply to it! It would be a near hopeless task. Camera based techniques do not require such re-engineering of the environment.

A question that arises is why is there no product of this type in the market? Is it because of legal worries – a guy land up in an accident and file a suit saying that a product weakness caused the accident. I don’t think that this is a major issue. Proper labeling and use of Terms and Conditions approved by a lawyer would safeguard the tool. Early versions of the product can be identified as being purely advisory and the user can be cautioned not to depend upon their accuracy and completeness in recognizing signs.  

An important research goal is to use two or more independent recognition techniques in one device and see how the results can be used together to improve product accuracy and reliability. Work on a prototype should be treated as a software engineering project, in which product reliability and performance are given importance and bugs are meticulously tracked and fixed. Optimizing the code to achieve quick response is obviously important too, but nothing can substitute for a well-thought out clean design to start with.

A camera based recognizer can be built as an app for a cell phone in the first place. This will also give it access to GPS coordinates of the vehicle it is mounted on and Web-based access to a remote server. It is one thing to recognize a stop sign at an intersection; it is far better to anticipate that traffic signal before the car comes within sight of that intersection. Integration of a map on the server with the traffic sign detector and recognizer will advance the state of the art to a superior level as compared to that of an unconnected sign recognizer.

Srinivasan Ramani