Z-ENG: Implementation of Multi-Object Real-Time Association and Tracking Algorithms
2023-2024 tavasz
Nincs megadva
Téma leírása
Autonomous transportation systems are designed to simultaneously recognize and track both (pseudo-)static and dynamic objects in their environment. During the digital 3D reconstruction of the environment, sensor fusion algorithms abstract and map the information collected from traffic participants into a model space, encompassing attributes such as the speed, size, position, orientation, and expected motion state of passing vehicles.
In real-world traffic scenarios, determining the motion states of multiple traffic participants simultaneously is crucial, often involving intersections, while dealing with significant measurement noise. In this context, numerous hypotheses need to be examined, and a real-time abstract representation of the environment must be constructed based on motion states, sensor-derived noise models, and estimated covariance values.
The student's task involves the implementation and quantitative evaluation of a multi-object real-time association algorithm capable of tracking multiple objects over time, associating them across successive timestamps, and filtering noisy measurement data, all derived from a real traffic dataset. The suggested algorithms are:
- Joint Probabilistic Data Association Filter
- Multiple Hypothesis Filter
For this task, the use of publicly available datasets is recommended, and for evaluation, reference data from these datasets should be used. Quantitative analysis will require the implementation and application of metrics such as MOTA (Multiple Object Tracking Accuracy) and MOTP (Multiple Object Tracking Precision).
To solve this task, the student will receive assistance from the employees of the Continental AI Development Center.
If you are interested in the topic, be sure to contact Dávid Sik by email before applying, indicating the selected topic, training level, major and the planned project subject.
Autonomous transportation systems are designed to simultaneously recognize and track both (pseudo-)static and dynamic objects in their environment. During the digital 3D reconstruction of the environment, sensor fusion algorithms abstract and map the information collected from traffic participants into a model space, encompassing attributes such as the speed, size, position, orientation, and expected motion state of passing vehicles.
In real-world traffic scenarios, determining the motion states of multiple traffic participants simultaneously is crucial, often involving intersections, while dealing with significant measurement noise. In this context, numerous hypotheses need to be examined, and a real-time abstract representation of the environment must be constructed based on motion states, sensor-derived noise models, and estimated covariance values.
The student's task involves the implementation and quantitative evaluation of a multi-object real-time association algorithm capable of tracking multiple objects over time, associating them across successive timestamps, and filtering noisy measurement data, all derived from a real traffic dataset. The suggested algorithms are:
- Joint Probabilistic Data Association Filter
- Multiple Hypothesis Filter
For this task, the use of publicly available datasets is recommended, and for evaluation, reference data from these datasets should be used. Quantitative analysis will require the implementation and application of metrics such as MOTA (Multiple Object Tracking Accuracy) and MOTP (Multiple Object Tracking Precision).
To solve this task, the student will receive assistance from the employees of the Continental AI Development Center.
Külső partner: Continental Autonomous Mobility Hungary
Maximális létszám:
1 fő