A python implementation of multi-model estimation algorithm, based on paper: research project: trajectory predication technology for autonomous driving vehicle
Mainly refer to paper:
- A comparative study of multiple-model algorithms for maneuvering target tracking
- Joint Monocular 3D Vehicle Detection and Tracking
- A Baseline for 3D Multi-Object Tracking
In autonomous driving systems, the prediction of the other vehicles' behavior in the next few seconds based on their current state, including position, velocity, and yaw angle, has a critical influence on ego-vehicle's decision on the behavior at the next moment.
Therefore, this project aims to implement a trajectory predictor, based on lidar and camera sensor. The result shows, the AMM algorithm based on the motion model can predict the future trajectory of the object within a certain range
-
We used the KITTI object tracking testset as dataset.
-
We used the camera and lidar as sensors, to get the position information of vehicle.
-
We used the Mono3DT(for camera) and AB3DMOT(for lidar) as detection and tracking model. A comparison of tracking results shows as following (left: lidar vs right: camera):
-
We used the AMM algorithm based on the motion model as prediction algorithm
-
Comparison of predicted result and ground truth at frame 0 (top left), frame 5 (top right), frame 10 (bottom left) and frame 20 (bottom right). The contour presents the probability distribution :
