Sensor and sensor Integration

1. Object detection and navigation of RC car suing vision and LIDAR integration on CAN bus

This result and research were published in the ISTAI 2018.

The primary objective of this project to demonstrate and test the capabilities of the LIDAR and Camera integration using the CAN bus.

The project was partially funded by kvaser.

In this project, the socket CAN network was constructed using Arduino and MCP CAN 2515 module. The controller, Sensor, and the actuation system which includes separate steering and traction (throttle) all are connected through this can communication. The CAN is used as a high-speed communication protocol which lets different CAN module interact with each other where priority between the is nodes is set using the arbitration introduced by the tax ID.

The sensor has the lower priority compared to the steering and throttle so it will have a higher taxing ID. The sensors used in this project are LIDAR and Camera.

The camera is a color code detection smart camera which will pass the bounding box values to Arduino which will in turn transfer the data to the CAN. A similar approach is used to the LIDAR.

Finally, data is fused, temporal alignment is done on the data and PID control logic is applied.

Personal Contribution: In this project, my contribution was related to vision and control design, sensor integration, temporal alignment of the sensor signal and actuation system. I was also responsible for setting up the CAN communication.

Future scope: This project is to elaborate and explore the use of CAN bus and it's used in the autonomous car. More experiment is needed to test the real-time speed of the CAN communication and data lost in the arbitration.

2. Training and Implementation of Haar Cascade using OpenCV for real time bject detection using Camera.

This was a summer project to learn and implement OpenCV and do necessary image processing. During this project, I was able to use the pre-made cascade in real time object detection. But cascades for a particular object was not available, I set my self to train a specialized cascade for football.

Here for the negative images, I scraped the internet using the python. I used OpenCV tools to generate the position images using a single positive model. Next step train the data for 20 stages to obtain the XML file of the cascade which can be used in real time object detection.

GitHub LInk is here