objectdetection

Documentation

Overview

Package objectdetection defines a functional way to create object detection pipelines by feeding in images from a gostream.VideoSource source.

Index

• func ConnectedComponents(img image.Image, isValid func(image.Image, image.Point) bool) ([][]image.Point, []image.Rectangle)
• func Overlay(img image.Image, dets []Detection) (image.Image, error)
• func OverlayText(img image.Image, text string) image.Image
• type ColorDetectorConfig
• type Detection
  • func NewDetection(boundingBox image.Rectangle, score float64, label string) Detection
• type Detector
  • func Build(prep Preprocessor, det Detector, post Postprocessor) (Detector, error)
  • func NewColorDetector(cfg *ColorDetectorConfig) (Detector, error)
• type Postprocessor
  • func NewAreaFilter(area int) Postprocessor
  • func NewLabelConfidenceFilter(labels map[string]float64) Postprocessor
  • func NewLabelFilter(labels map[string]interface{}) Postprocessor
  • func NewScoreFilter(conf float64) Postprocessor
  • func SortByArea() Postprocessor
• type Preprocessor
  • func ComposePreprocessors(pSlice []Preprocessor) Preprocessor
  • func RemoveColorChannel(col string) (Preprocessor, error)

Functions

func ConnectedComponents

func ConnectedComponents(img image.Image, isValid func(image.Image, image.Point) bool) ([][]image.Point, []image.Rectangle)

ConnectedComponents takes in an image frame and a function that determines if a point in the image is valid and returns two slices. The first returned slice is a 2D slice of Points, with each 1D slice corresponding to a contiguous set of valid Points within the image. The second returned slice contains Rectangles, each of which constitutes the bounding box for the set of valid Points at the same index.

func Overlay

func Overlay(img image.Image, dets []Detection) (image.Image, error)

Overlay returns a color image with the bounding boxes overlaid on the original image.

func OverlayText

func OverlayText(img image.Image, text string) image.Image

OverlayText writes a string in the top of the image.

Types

type ColorDetectorConfig

type ColorDetectorConfig struct {
	resource.TriviallyValidateConfig
	SegmentSize       int     `json:"segment_size_px"`
	HueTolerance      float64 `json:"hue_tolerance_pct"`
	SaturationCutoff  float64 `json:"saturation_cutoff_pct,omitempty"`
	ValueCutoff       float64 `json:"value_cutoff_pct,omitempty"`
	DetectColorString string  `json:"detect_color"` // hex string "#RRGGBB"
	Label             string  `json:"label,omitempty"`
}

ColorDetectorConfig specifies the fields necessary for creating a color detector.

type Detection

type Detection interface {
	BoundingBox() *image.Rectangle
	Score() float64
	Label() string
}

Detection returns a bounding box around the object and a confidence score of the detection.

func NewDetection

func NewDetection(boundingBox image.Rectangle, score float64, label string) Detection

NewDetection creates a simple 2D detection.

type Detector

type Detector func(context.Context, image.Image) ([]Detection, error)

Detector returns a slice of object detections from an input image.

func Build

func Build(prep Preprocessor, det Detector, post Postprocessor) (Detector, error)

Build zips up a preprocessor-detector-postprocessor stream into a detector.

func NewColorDetector

func NewColorDetector(cfg *ColorDetectorConfig) (Detector, error)

NewColorDetector is a detector that identifies objects based on color. It takes in a hue value between 0 and 360, and then defines a valid range around the hue of that color based on the tolerance. The color is considered valid if the pixel is between (hue - tol) <= color <= (hue + tol) and if the saturation and value level are above their cutoff points.

type Postprocessor

type Postprocessor func([]Detection) []Detection

Postprocessor defines a function that filters/modifies on an incoming array of Detections.

func NewAreaFilter

func NewAreaFilter(area int) Postprocessor

NewAreaFilter returns a function that filters out detections below a certain area.

func NewLabelConfidenceFilter

func NewLabelConfidenceFilter(labels map[string]float64) Postprocessor

NewLabelConfidenceFilter returns a function that filters out detections based on label map. Does not filter when input is empty.

func NewLabelFilter

func NewLabelFilter(labels map[string]interface{}) Postprocessor

NewLabelFilter returns a function that filters out detections without one of the chosen labels. Does not filter when input is empty.

func NewScoreFilter

func NewScoreFilter(conf float64) Postprocessor

NewScoreFilter returns a function that filters out detections below a certain confidence.

func SortByArea

func SortByArea() Postprocessor

SortByArea returns a function that sorts the list of detections by area (largest first).

type Preprocessor

type Preprocessor func(image.Image) image.Image

Preprocessor will apply processing to an input image before feeding it into the detector.

func ComposePreprocessors

func ComposePreprocessors(pSlice []Preprocessor) Preprocessor

ComposePreprocessors takes in a slice of Preprocessors and returns one Preprocessor function.

func RemoveColorChannel

func RemoveColorChannel(col string) (Preprocessor, error)

RemoveColorChannel will set the requested channel color to 0 in every picture. only "R", "G", and "B" are allowed.