Pedestrian and Vehicle Detection Using Night-Vision Camera through CNN on Indian Roads
Mayank Raj, Swet Chandan · 2018 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN) · 2018
Low visibility is one of the leading causes of vehicle accidents on Indian roads. Presently there are not many research accessible to manage these sorts of circumstances. Performance of object detection algorithm has very less accuracy in case of night time because the intensity of luminance at night is very less with respect to daytime, even human eye unfit to foresee all objects at night. For better accuracy of detection, we have to use thermal vision camera which costs a lot. In the present work we proposed a Convolutional Neural Network (CNN) based modified Single Shot Multi-Box Detection (SSD) method to identify the pedestrian and vehicles at night time utilizing night-vision camera. We have tried the actualized calculation on tests from Delhi-NCR area which incorporates recordings of highways and road streets. We have implemented certain filters as pre-treatment of sample videos (29fps, 1080p) before implementing the algorithm to improve precision. With the help of our modified algorithm, we are able to detect vehicle and pedestrian using the night-vision camera in real-time. We have achieved an accuracy of 85.28% which is superior to any other algorithm and process in this field.