Projects

3D Object Detection

• End-to-end LiDAR-based detector
• Real-time visualization with predicted bounding boxes
• Multi-threaded pipeline for data loading, inference and display

3D Detection & Tracking on Edge

• Fusion of LiDAR and camera data
• High-performance detection on edge devices
• Real-time SORT tracking with remote monitoring via websockets

3D Point Cloud Scene Reconstruction

• Unified interface for NuScenes, DDAD and KITTI
• Self-supervised learning without ground-truth depth
• Integrated 3D visualization with CamViz

2D Panoptic Segmentation

• Real-time inference on PC and edge devices
• Multi-threaded architecture with teacher–student distillation
• Prototype web app built with Gradio

2D Object Detection

• Fast, lightweight model on Jetson edge devices
• Real-time display with minimal latency
• Web prototype with Gradio

Multi-Camera Human Intrusion Detection

• Multi-camera surveillance for industrial safety
• Real-time instance segmentation
• Multi-threading and multi-processing pipeline

3D Object Detection with Spatial Shape Transformer

A Transformer-based 3D detector that models object dimensions and spatial relationships for accurate LiDAR detection.

3D Detector for Occluded Objects

A point–voxel framework that produces high-quality 3D boxes for occluded objects while preserving shape context.

Improving Object Shape in 3D Detection

Improves object-shape accuracy of 3D detectors with a lightweight, quantized model.

Jumping Optimization for Quadruped Robots

Optimizing jump trajectories for the Unitree A1 quadruped in MATLAB and ROS/Gazebo.