TrackKLTOptions Struct
Pyramidal Lucas-Kanade / KLT tracker options. More...
Declaration
Included Headers
Public Member Functions Index
| std::string | summary () const |
|
Human-readable, non-throwing summary for logs, diagnostics, and Python repr. More... | |
Public Member Attributes Index
| int | width = 0 |
| int | height = 0 |
| int | batch_size = 1 |
|
Number of packed image/point sets in one dispatch. Public tensor shape uses this as N. More... | |
| int | num_points = 0 |
|
Number of input points tracked per batch item. More... | |
| int | win_half = 10 |
|
Half-window radius for LK tracking. Must fit the EV74 graph envelope. More... | |
| int | max_iters = 30 |
|
Maximum LK solver iterations per pyramid level. Must be positive. More... | |
| int | max_level = 3 |
|
Maximum pyramid level. Must fit the EV74 graph envelope. More... | |
| int | detect_new_features = 0 |
|
When non-zero, publish detected replacement/new features as a third output. More... | |
| int | fast_threshold = 30 |
|
FAST threshold used only by the detect-new-features path. More... | |
| int | max_features = 500 |
|
Maximum detected replacement/new features per batch item. More... | |
| int | grid_x = 8 |
|
Horizontal grid cells for detect-new-features distribution. More... | |
| int | grid_y = 6 |
|
Vertical grid cells for detect-new-features distribution. More... | |
| int | min_px_dist = 10 |
|
Minimum pixel distance for detect-new-features. Must be non-negative. More... | |
| int | debug = 0 |
|
EV graph debug level. Current native visual graphs accept values in [0,2]. More... | |
| int | num_buffers = 0 |
|
Optional processcvu queue/buffer override. 0 keeps the plugin/runtime default. More... | |
| std::string | element_name |
|
Optional GStreamer/processcvu element name. Empty means Neat generates a stable name. More... | |
| std::string | prev_image_name = "prev_image" |
|
Previous grayscale image tensor name. More... | |
| std::string | cur_image_name = "cur_image" |
|
Current grayscale image tensor name. More... | |
| std::string | input_points_name = "input_points" |
| std::string | output_points_name = "output_points" |
| std::string | output_status_name = "output_status" |
| std::string | output_features_name = "output_features" |
|
Optional published feature-list tensor name when detect_new_features != 0. More... | |
| bool | publish_error = false |
|
Append mean absolute photometric residuals at the returned coordinates. More... | |
| std::string | output_error_name = "output_error" |
|
Residual output name. Unevaluated residuals are positive infinity. More... | |
| float | error_threshold = -1.0f |
|
Maximum mean absolute UInt8 intensity residual [0,255]; -1 disables filtering. More... | |
| float | min_eig_threshold = 0.0f |
|
Minimum texture eigenvalue divided by window area and 255 squared; 0 disables. More... | |
| bool | require_convergence = false |
|
Require convergence of the final full-resolution solve, not every coarse level. More... | |