Smart Traffic Signalling Using Computer Vision for Avoidance of Blind Spots

Abstract

A blind spot is the area or zone on the road outside the driver’s field of vision. Essentially, it is an area that cannot be seen by looking forward through your windscreen, or by using your rear-view and side-view mirrors. It is hidden by parts of the vehicle’s structure. Blind spots can be large enough in size to easily block another car, motorbike, cyclist, or pedestrian from your view. By the time a driver realizes that another vehicle is approaching from the other end it might be too late to control the vehicles. It is found that many critical accidents are caused by this situation. The proposed system is built to avoid accidents at blind spots using TensorFlow (Frozen Model) for the database and some OpenCV libraries for recognizing the objects. The model imported for detection consists of classified 80 objects, of which only 4 objects (Car, Bikes, Trucks and Buses) have been chosen. 80 classes labels are stored  as a list to refer to a particular object. Using the previous second’s data analysis, the current situation is interpreted. Depending on the increasing or decreasing area of the vehicle in the camera frame, a decision is taken whether the vehicle is approaching or moving away. This data is used to give signals on both roads for safe passing. Open source Flask software is used in the backend to store this real time data in the form of graphs. The accuracy of the model was found to be 76.9%. This model is successfully able to detect vehicles on the blind turn roads. This can be employed in our real-world applications in roads. They can help reduce accidents significantly. All the data collected by the camera regarding the number of vehicles passed, time of passing etc. can be stored in the database(backend) for the developer of the traffic management system to decide when the model should be very active and times when the vehicle detection by the model can be turned off. The data stored is the number of vehicles passed in a road and the timestamp of each vehicle passing. The visualization of number of vehicles on the road Vs time is also performed for better understanding of the road traffic situations.