Robust detection of non-motorized road users using deep learning on optical and LIDAR data

Detection of non-motorized road users, such as cyclists and pedestrians, is a challenging problem in collision warning/collision avoidance (CW/CA) systems as direct information (e.g. location, speed, and class) cannot be obtained from such users. In this paper, we propose a fusion of LIDAR data and a deep learning-based computer vision algorithm, to substantially improve the detection of regions of interest (ROIs) and subsequent identification of road users. Experimental results on the KITTI object detection benchmark quantify the effectiveness of incorporating LIDAR data with region-based deep convolutional networks. Thus our work provides another step towards the goal of designing safe and smart transportation systems of the future.