ovl 2020.1.1

A modular and versatile Python package for Computer Vision object detection pipelines Tailored for robotics applications

Ovl-Python

Python Module for Computer Vision Object Tracking and Detection mainly for the FIRSTֲ® Robotics Competition Program

There have been significant changes from the previous version - changelog

Dependencies:

The following python module dependencies are needed:

• OpenCV

• numpy

The following python modules are optional for certain features:

• NetworkTables (pyNetworkTables) for: NetworkTableConnection

• sklearn (scikit-learn): HSVCalibration

OVL is officially supported for python 3.5+!

Installation:

Using pip:
python -m pip install ovl

Usage:

The library uses simple yet highly customizable syntax to create a vision pipeline using the Vision object

A pipeline that detects a yellow circle:

import ovl

# filter contours that are larger than 200 pixels
# and are approximately a circle and then sort by size

contour_filters = [ovl.area_filter(min_area=200), ovl.circle_filter(min_area_ratio=0.7), ovl.dec_area_sort()] 

threshold = ovl.YELLOW_HSV # Define the wanted color to detect 
camera = ovl.Camera(0) # open the first connected camera 

yellow_circle = Vision(threshold=threshold,
                       contour_filters=contour_filters,
                       camera=camera,
                       image_filters=image_filters)

while True:
    image = yellow_circle.get__filtered_image()
    contours, image = yellow_circle.detect(image)
    directions = yellow_circle.get_directions(contours, image)
    
    print(directions) # prints out the (x, y) coordinates of the largest target

Documentation

The Hebrew and English documentation have been removed all functions and classes now have in code documentation. This includes code examples, recommended usage reference to further documentation and more.

simply use help on a class/function:

help(ovl.Vision)

help(ovl.contour_filter)

help(ovl.area_filter)

Major changes from previous patch:

Dependency changes:

Now only CV2 and numpy are mandatory dependencies all other are optional:

• sklearn is only needed for calibration
• networktables is only needed for NetworkTableConnection
• PySerial is only needed for SerialConnection

Additions:

The new version has brought many new additions

Vision

Major Vision object overhaul:

• removed many redundant functions
• removed obsolete parameters, cleaned overall usage
• extracted network settings to the new Connection object
• Constructor has been rewritten to be quicker and less complex
• direction functions and target amount have been extracted to new Director object
• Vision.get_contours -> Vision.find_contours
• Vision.get_contours_mask -> Vision.find_contours_in_mask

Other Vision related additions

• New Vision-like, AmbientVision, allows 2 Visions to be run at once by taking "turns"
• New Vision-like MultiVision, allows of multiple visions to be run and swapped, supports AmbientVision

Connection

• New object Connection, represents a connection
• New connection NetworkTableConnection
• New connection SerialConnection

Threshold

New functions for displaying images: stitch_images, show_image New image filter crop_image, can crop a rectangle out of an image New object Threshold, converts an image to a binary image New Threshold object CannyEdge New Direction Utility CameraSettings, defines camera calibrations, offset and other unique changes New object CameraCalibration, can be used to Calibrate an camera New object DirectionMonitor, Alters directions according to various situations

Images and Utilities

• photo_array extracted from Vision, now a standalone utility function
• Naming conventions for photo array are now functions, custom ones can be created

Contour filters

Now Filters can be preloaded with their parameters using the matching XXXX_filter Decorator:

• New Filter Decorator, contour_filter turns a function to a contour filter that iterates on a list of contours
• New Filter Decorator, conditional_contour_filter turns a function that filters one contour to a contour filter
• New Filter Decorator, image_filter turns a function to a image filter that modifies the given image somehow
• New contour filter, circle_filter, detects simple circles based on fill ratio

Color and MultiColor

• Color and MultiColor now inherit from Threshold
• Color has been rewritten for a cleaner usage

Director:

New object Director, Handles directing from a final list of targets (contours):

• Receives a directing function that receives a list of contours and the image
• Receives a list of DirectionMonitors that alter the directions
• The directing function returns the initial direction
• The list of Direction monitors are then applied on the result which is returned

Other changes:

• Camera object now automatically starts
• BuiltInColors no longer a class
• serializing all objects have been temporarily removed, they will be re-added in an upcoming patch

Renames:

• circle_filter -> constraining_circle_filter
• Calibration -> HSVCalibration
• get_fill_ratio_triangle -> triangle_fill_ratio
• get_fill_ratio_straight -> rectangle_fill_ratio
• get_fill_ratio_rotating -> rotating_rectangle_fill_ratio
• get_fill_ratio_circle -> circle_fill_ratio
• get_contour_center -> contour_center
• get_approximation -> contour_approximation
• get_lengths_and_angles -> contour_lengths_and_angles
• n_sided_polygon_angle -> regular_polygon_angle

Removals:

• ToolClass
• ToolError

Upcoming Changes and Additions

• Serialization and Deserialization to json of all objects
• CompoundVision, a vision like object that can has multiple changeable configurations
• Filter flow control, for complex logical combinations of contour filters and conditional contour filters
• HSVCalibration overhaul
• OVLRaspbian, a Raspbian based image for easy strap-booting on RaspberryPI
• Graphing tools for data analysis of performance and function
• Live Graphing and Graph Streaming for Graph Capabilities
• A debugger tool for tuning Vision pipelines and detections for ease of use and work efficiency

Have a feature you are interested in? Consider suggesting it or contributing!