TASI Research
will concentrate on the design and development of a e-scooter mounted data collection system and an offline data processing pipeline. It will focus specifically on data collection to aid scenario reconstruction of the events and accurate tracking of e-scooters with respect to static coordinate frames such as on the road and moving coordinate frames such as on the cars on the road.
The data collection system will aim at a design to fit the requirements and constraints of an e-scooter. The design will be geared towards having features such as online synchronization and compression of data coming from the monocular cameras, the LIDAR and the IMU in order to aid the processing stages. The system utilizes a Jetson TX2 to aid in the storage of the recording of the sensor data from Ouster LIDAR and the FLIR and Webcam cameras to collect and store data. ROS drivers for Network Packet Capture of the LIDAR data and and Gstreamer with NVIDIA's hardware accelerated image conversion is utilized to aid in data storage.
The processing pipeline divided into five parts. The first part of the offline processing pipeline deals with the unpacking and formatting of the sensor data and the synchronization data for further processing.
The second part will deal with the filtering and fusing data from the camera and the LIDAR to solve the DATMO problem and effectively detect and track the car in question The object detection from the camera will primarily use segmentation to aid detection. The calibration of the intrinsic and extrinsic matrices will allow for accurate fusion mapping with the lidar point cloud and help map relative trajectories of the e-scooter and then use the data to understand the relative trajectories of the e-scooter with respect to the car coordinate frame.
The third part of the processing pipeline then deals with fusing the data from the LIDAR and IMU to solve the SLAM problem to help map the static environment with the use of the Google Cartographer software as well as provide the trajectories of the e-scooter w.r.t the static environment.
The final part of the processing pipeline will then deal with piecing data from SLAM and DATMO to recreate the scenarios and map out the trajectories with respect to the road or world coordinate frame.
Data Collection and Processing Pipeline
3D LIDAR-Camera Sensor Fusion