Multi-Camera Fusion in Apollo Software Distribution
| Autor: | Aleksandr Buyval, Geesara Kulathunga, Aleksandr Klimchik |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
0209 industrial biotechnology
Computer science business.industry 020208 electrical & electronic engineering 02 engineering and technology Kalman filter Software distribution Tracking (particle physics) Object (computer science) Object detection Visualization 020901 industrial engineering & automation Control and Systems Engineering Video tracking Obstacle 0202 electrical engineering electronic engineering information engineering Computer vision Artificial intelligence business |
| Zdroj: | IFAC-PapersOnLine. 52:49-54 |
| ISSN: | 2405-8963 |
| DOI: | 10.1016/j.ifacol.2019.08.047 |
| Popis: | We have continued extending the Apollo software distribution for supporting multi-object tracking followed by the fusion in a real-time manner for a multi-camera setup. Our proposed multi-camera fusion approach includes improvements in the following stages: object detection, data association, filtering and tracking on each camera space, and adding support for connecting multiple cameras into the system. Object detection and data association are the pivots for object tracking across the multi-camera setup. Hence, YOLOv3 that is the state-of-art model for real-time object detection is introduced to improve accuracy. When the obstacle sizes are changing over time abruptly, most of the popular data association approaches do not perform well. A new distance estimation technique is purposed to overcome this difficulty. Kalman filter that uses for tracking objects in 3D has modified to improve the estimation accuracy of pose and distance to the object from the car. Subsequently, added support for multi-camera visualization as well. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|