How reliable is pedestrian detection for autonomous vehicles ?
With growing interest in autonomous driving applications, there is an ever-growing interest in pedestrian detection techniques. While advances in deep learning techniques have stepped up detection performance over the past years, there is quite a road to cover. Modern pedestrian detection systems often perform sub-optimally in practical deployment scenarios (e.g : autonomous vehicles). This sets an upper limit on the safety of modern autonomous driving industry. In this talk, we discuss 2 research directions -- utilization of multiple layers simultaneously and explicitly reducing the impact of background noise on the detection performance. These research directions bring out critical research questions in machine learning and computer vision which can have far-reaching consequences for all object detection techniques.