Machine perception

Project : The goal of this project was to develop an algorithm which was able to detect a pedestrian in an image and localize the pedestrian in the world frame. Using collected sensory data from the following sensors: LIDAR, stereo camera and mono camera. 

Solution: In our approach, we utilized a pretrained CNN to detect pedestrians from preprocessed images captured by a single camera. For precise localization, we combined data from LIDAR and a camera, which proved to be the most accurate. Additionally, we employed stereo vision using two cameras to gauge distance, and mono vision, where we projected a line from a single camera to determine its intersection with a plane for estimating the pedestrian's location. The LIDAR and camera fusion method was the most effective in accurately localizing the pedestrian.