Low Light Object Detection - Intel Advanced Driver Assistance Systems Project (2019)

Final Presentation

One Sentence Summary

Performed Semantic Segmentation on the Indian Driving Dataset using custom trained DeepLabV3+ network and worked on low light enhancement techniques to improve the model’s accuracy.

Problem Statement

To enhance the vision of the driver with smart algorithms that can work in low light scenarios – dusk, night, in lights from headlights etc. The solution should be able to detect common objects like cars, two-wheelers, auto rickshaws, pedestrians, riders in the night/low-light conditions. The system has been developed to work in the visible spectrum.

Comparision of Different Low Light Enhancement Techniques

1. Histogram Equalization
2. An enhancement algorithm using Exposure Fusion Framework
3. Retinex based end to end deep learning model for image enhancement
   

Awarded Best Completed Submission winner in the Intel - PESU Student Contest among 70 teams. I took part in this competition as a first year undergraduate student and was the youngest participant in the contest.