A Real Time Object Detection, Classification and Counting of Bengaluru Traffic Surveillance Using AI and ML

Abstract: Object Detection and tracking of objects is a field that has many applications in this rapidly developing society with more cameras being set up all over the world. Traffic surveillance has become the most pressing issue in increasingly developing cities. Due to poor traffic management in the city of Bangalore, a lot of manpower and hours are being used up. Our project provides a system that detects and monitors vehicles, pedestrians, traffic signals, and signboards and keeps a count of the number of objects per class passing through. This is built using a custom YOLOv4 dataset and functions pertaining to Bengaluru Traffic and implemented using YOLOv4 and Tensorflow. Our model managed to raise the number of objects detected by over 60% to 94% but went down in predicting accuracy from 90% to 65% compared to foreign datasets.

Keywords: YOLOv4, Tensorflow, Non-max Suppression, Darknet.