Image Feature Point Separation for Accurate Object Recognition

The patent segments feature points into two distinct categories: object feature points and distant scene feature points. This segmentation is achieved by classifying line segments based on their motion characteristics - object line segments exhibit significant positional changes between frames, while distant scene line segments show minimal movement. By separating these categories, the system maintains a large quantity of feature points for processing while preventing erroneous associations between object and background features, thus preserving recognition accuracy.

The patent extracts and isolates object feature points from the mixed set of feature points by identifying and removing distant scene feature points. This is accomplished by comparing line segment positions across consecutive frames and filtering out those with minimal displacement, which are characteristic of distant background elements. The extraction process retains only the moving object feature points for association and recognition, eliminating the harmful influence of stationary background features while maintaining sufficient feature point quantity for robust recognition.

The patent applies local quality by treating different regions of the image differently based on their motion characteristics. Instead of uniformly processing all line segments, the system identifies local regions corresponding to moving objects versus stationary distant scenes by analyzing positional changes. Object line segments in moving regions are processed with one set of rules (expecting significant displacement), while distant scene line segments in stationary regions are processed with different rules (expecting minimal displacement). This localized differentiation ensures accurate position tracking for objects while filtering out erroneous features from the distant scene.

The patent changes the parameter used for feature point selection from static geometric properties to dynamic motion-based properties. Instead of relying solely on geometric features like intersection points which remain stationary in the image coordinate system, the system uses positional variation between frames as the key parameter. Line segments are classified and feature points are selected based on whether their position changes significantly between consecutive frames, transforming the selection criterion from a static geometric parameter to a dynamic motion parameter that inherently distinguishes objects from background.