Drivable Area Detection Using Corrected Obstacle Trajectories

The patent applies preliminary action by correcting historical trajectories and position information before using them for drivable area detection. The system pre-processes detection data from multiple frames, adjusting trajectories of dynamic obstacles and positions of static obstacles based on map information and key point comparisons. This preliminary correction ensures that the data used for boundary determination is already optimized, improving detection accuracy without adding complexity during the main detection process.

The patent implements feedback by using map information and key point comparisons to verify and correct detected positions and trajectories. The system continuously compares detected key points against map data and historical information, making iterative corrections to trajectory and position data. This feedback mechanism ensures that detection results converge toward accurate values, resolving the contradiction between simplicity and precision.

The patent applies segmentation by dividing the correction process into discrete key points along trajectories and boundaries. Instead of correcting entire continuous paths at once, the system identifies and corrects specific key points (such as intersection points, turning points, and boundary markers) independently. This segmentation allows parallel processing and reduces computational complexity while maintaining boundary smoothness, as each key point can be corrected using local map information without affecting the entire trajectory.

The patent implements local quality by applying different correction strategies to different parts of the detection data based on their specific characteristics. Static obstacles near key points receive correction based on map information, while dynamic obstacles have their trajectories corrected using a combination of historical data and current frame information. The system adjusts the degree of correction locally, applying stronger constraints in areas requiring high precision (such as boundary regions) and more flexible correction in open areas, thereby optimizing processing efficiency.

The patent applies preliminary action by pre-processing multiple video frames to extract and store trajectory information before the main detection process. The system analyzes historical frames in advance, establishing baseline trajectories and key point positions that can be referenced during real-time detection. This pre-extraction of trajectory data from multiple frames reduces the computational burden during actual drivable area calculation, as the system only needs to refine rather than compute trajectories from scratch.

The patent implements partial action by selectively processing only the necessary portions of multiple video frames. Instead of analyzing every pixel and object in all historical frames, the system focuses on extracting trajectories of relevant dynamic obstacles and positions of static obstacles near key points. The system processes excessive data (multiple frames) but only performs detailed analysis on the partial subset that directly impacts drivable area boundaries, thereby maintaining trajectory accuracy while controlling processing complexity.