An Augmented Deep Learning Inference Approach of Vehicle Headlight Recognition for On-Road Vehicle Detection and Counting

Vehicles have been a big part of many lives; from the time it is invented and as it increases popularity in the 20th century. Though they offer the benefit of convenience, they also have certain negative effects as they add to air pollution and global warming, as well as risks when they are not handled properly. In recent years, the number of vehicles on the road is rapidly increasing and it causes different major concerns. One of the major effects of this increasing volume of vehicles is the traffic congestion it caused on our roads especially in the urban areas. This traffic congestion became one of the major problems in many cities in the world including Metro Manila, Philippines. Many options are discussed and implemented by the traffic management but it seems that it is still unsolved. In recent years, traffic congestions became unpredictable, there are parts of the cities that don’t experience traffic congestion then suddenly traffic builds up to that area. Also, traffic congestion might happen every hour of the day. With this concern, the study proposed a system for vehicle headlight recognition for on-road vehicle detection and counting. This study focused on the detection of the headlight of every vehicle that will be seen on the perimeter of the installed camera. The system can detect headlight vehicles during daytime and nighttime as we trained the AI to recognized the headlights in these scenarios. Possible applications of this system can be in monitoring the volume of vehicles within the area and it can be used by traffic management authority in monitoring the build-up of traffic or traffic situation in an area so that they can provide an immediate solution, such as re-routing, one-way street conversions, etc.