Investigation into Raindrop Detection and Removal for Improved Sensing in Automotive Computer Vision Applications
This event took place on Wednesday 18 December 2013 at 11:30
Automotive Computer Vision is becoming more mainstream as the hardware for such is becoming more accessible. This is giving rise to implementations being added to such systems, of which most are indended to increase safety. Examples include Driver Drowsiness Detection, Headlamp detection, Speed limit notification and Predictive Emergency Braking. Raindrops however can distort the image, drematically decreasing the accuracy of the information supplied to the driver. Within the work carried out, we implement a current technique to detect raindrops through saliency and adapt the verification phase to improve accuracy of classification using various machine learning algothrims. The preformance of these results are analysed to determine the best approach. We then investigate how raindrops degrade the results produced from a typical automotive computer vision task and if the technique of stereo infilling or impainting yeald better results.
Watch the webcast replay >>