> For the complete documentation index, see [llms.txt](https://docs.augelab.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.augelab.com/function-blocks/blocks-reference/detections-shapes/detectors/feature-detector.md).
# Feature Detector
This function block finds an object in a live or static image by matching features from a provided training image. It compares distinctive points between the training image and the input image, attempts to estimate the object's position and rotation, and provides a visual and numeric result.
## đĨ Inputs
`Train Image` The reference image that contains the object to detect. Provide a clear, well-lit crop of the object.
`Input Image From Camera` The image where the block will search for the object (live camera frame or loaded image).
These are input sockets.
## đ¤ Outputs
`Detected Image` The input image annotated with detection graphics (e.g., contour and center) when the object is found.
`Detect Status` A boolean indicating whether the object was detected (true/false).
`Center` The (x, y) coordinates of the detected object's center.
These are output sockets.
## đšī¸ Controls
`Homography Type` Choose the method used for estimating the geometric transform between images.
`Compute Type` Select the internal computation preference (quality vs performance).
`Number of features` How many keypoints/features to detect in the training image.
`Distance Threshold` Tolerance used when matching features (affects match acceptance).
`K nearest` Number of nearest matches considered when comparing descriptors.
`Pyramid Decrease Ratio` Controls scale down steps for multi-scale detection.
`Pyramid Levels` Number of pyramid levels used in multi-scale processing.
`Point Compare Type` Internal parameter affecting how points are compared.
These are widgets you can tune to adjust detection behavior.
## âī¸ How it runs
* The block extracts features from the provided `Train Image` using the chosen feature/computation settings.
* It then searches the `Input Image From Camera` for matching features and attempts to estimate the object's location and orientation.
* If a valid transformation is found, the block annotates the input image with a bounding contour and a center point, sets `Detect Status` to true, and outputs the detected center coordinates.
* If no valid match is found, `Detect Status` is false and `Center` is returned as a default coordinate.
The block continuously evaluates when connected to live sources, so tuning controls will update behavior in near real time.
## ⨠Key features
* Robust feature-based detection that handles rotation and scale changes.
* Multi-parameter tuning for quality vs performance trade-offs.
* Visual feedback via an annotated `Detected Image` to confirm results.
* Outputs both a boolean status and exact center position for downstream logic.
## đ Usage instructions
1. Provide a clear cropped example of the object to `Train Image`. Prefer images with distinct textures or corners.
2. Connect a camera frame or image to `Input Image From Camera`.
3. Start with moderate values: set `Number of features` to a few hundred and `Distance Threshold` to a middle value.
4. Run the system and observe the `Detected Image`. Adjust `Homography Type`, `Number of features`, `Distance Threshold`, and `K nearest` to improve reliability.
5. Use the boolean `Detect Status` to trigger actions (logging, saving, counting) or feed the `Center` into other blocks.
## đĄ Tips and Tricks
* If object and camera scales differ, use `Image Resize` before this block so the training image and input image are closer in scale.
* Crop the search area with `Image ROI Select` to speed up detection and avoid false matches from cluttered backgrounds.
* Improve robustness under noisy conditions with `Blur` to remove high-frequency noise, or use `Contrast Optimization` to enhance contrast before detection.
* Use `Show Image` to preview detection results interactively while you tune controls.
* For drawing and reporting results, combine with `Draw Detections` to overlay detections cleanly on output images.
* If template-based matching is more suitable for your object, consider trying the `Find Object` block as an alternative or complementary approach.
* To store examples or results for offline review, pair with `Image Logger`.
(hints above refer to available function blocks you can combine in a flow.)
## đ ī¸ Troubleshooting
* No detection: increase `Number of features`, relax `Distance Threshold`, or provide a clearer training image. Try different `Homography Type` values.
* False positives: reduce `Number of features` or tighten `Distance Threshold`. Crop the image using `Image ROI Select` to eliminate confusing backgrounds.
* Slow performance: lower `Number of features`, reduce pyramid levels, or run a smaller search region. Preprocess with `Image Resize` to reduce input dimensions.
* Poor results under low contrast: apply `Contrast Optimization` first or capture a higher-quality training image.
If results remain inconsistent, iterate: adjust controls incrementally and use `Show Image` to visually confirm the changes.