Yunpeng Xu

This project was carried out in CyLab, Carnegie Mellon University and advised by Prof. Ragunathan “Raj” Rajkumar. The final objective is to build a system that can show the surround view of a vehicle and also detects moving objects based on the four fish-eye cameras on TI TDX2x and Nvidia TX1 platform.

Moving object detector - Nvidia TX1 platform

My role

• Implemented moving object detection with mono fisheye camera on Nvidia TX1 platform
• Evaluated the performance of OpenCV application and applied CUDA OpenCV to optimize the application
• Improved each frame’s processing time by 23.5% after optimization using CUDA

Details

Figure 1. MOD algorithms details based on OpenCV

Figure 2. MOD performance measure before optimization on Nvidia TX1 platform

Figure 3. MOD performance measure after optimization for the same frame on Nvidia TX1 platform

Test

Video 1. a brief video that is recorded using Quad-View on TI TDA2x platform and analyzed using Nvidia TX1 platform in the CMU garage [click the graph]

Video 2. a brief video that is recorded using Quad-View on TI TDA2x platform and analyzed using Nvidia TX1 platform on the highway from Waterfront to Downtown in Pittsburgh, PA, USA [click the graph]

Surround view monitoring - TI TDA2x platform

My role

• Ported the MOD algorithm to TI TDA2x and help bring up surround view monitoring system with TI platform
• Applied prototype to test the built system, and then deploy the system to real car for further evaluation
• Do calibration based on the positions of vehicle and mounted cameras and live demo the whole system

Details

Figure 4. TI TDX2x prototype

Calibration

We also verified the system we built using real-time recorded videos and evaluated the system performance. Figure 5. Surround View Calibration

Live Demo

Figure 6. Live Demo Environment

Figure 7. Prof. Raj Rajkumar, Albert Davies and Me (from right to left)

Video 3. A live demo video shows both MOD and Surround View algorithm integrated on the TI TDA2x platform [click the graph]