Real-Time Vehicle and Pedestrian Detection Through SSD in Indian Traffic Conditions
Mayank Raj, Swet Chandan · 2018 International Conference on Computing, Power and Communication Technologies (GUCON) · 2018
Object detection is a standout amongst the most vital segment of self-driving cars. It is very challenging especially in the case of Indian roads which possess numerous issues including non-standard vehicles, unskilled drivers, lower visibility due to pollution and despicable path division. In this paper, we have to test our algorithm for Indian traffic conditions through Single Shot Multi-box Detection (SSD) method. For this, we have run our algorithm on the live feed video captured by the camera (30fps) for 25 km stretch and approximately 8000 image datasets on the Noida-Greater Noida Expressway. The implemented algorithm furnishes an accuracy of 96% with a high obvious positive rate of 94.23% and a minor false positive of 4%.