Trajectory Prediction using Deep Learning With Steering & Speed Recommendation

This is part of my Master Thesis. You can follow my progress here: https://www.linkedin.com/in/sarimmehdi550/detail/recent-activity/shares/ Code: https://github.com/sarimmehdi/master_thesis The code involves a 3D object detector and a neural net for semantic segmentation of the road. Then, the 3D bounding box is projected forward in space in the heading direction and that's how the heatmap is generated. Speed and steering recommendation was just something extra I decided to implement to show the potential usefulness of heatmaps like these. Of course, a reinforcement learning agent could do this job much better. The dataset I used for this video is the Waymo Open Dataset: https://waymo.com/open/download/ Since a pre-trained neural net for 3D object detection on this dataset was not available, I decided to use the ground truth bounding boxes. You can visit my LinkedIn page and see the same algorithm implemented on the KITTI dataset but without using any ground truth bounding boxes. I plan on generating a depth map and converting that into a point cloud and using that for 3D object detection (Pseudo-LIDAR). The results will be interesting to see. Stay tuned! #deep