Shot Detection

AI Basketball Shot Detection and Analysis

This program is able to detect when a shot occurs and fill in the balls flight from captured data. It calculates the balls initial velocity and launch angle. It is able to estimate the balls flight perpedicular to the camera plane (The z axis) using a single camera. The program is also able to detect when the balls flight is interupted by another object and will drop those data points.

Tracking and analysis performed on

unstable video

Program output world coordinates:

Tracking and analysis performed on

shot interrupted by person

Program output world coordinates:

Tracking and analysis performed on

shot interrupted by object
shot angled with component perpendicular to the camera plane

Program output world coordinates:

Formats for representing video frame object bounding boxes detected

Object Line Format (OL) CSV file

Each object's bounding box in a video recoreded on an individual line

clip_ID	width	height	frame	category	score	x1	x2	y1	y2	model
int	int	int	int	string	float	int	int	int	int	string

Mock 1 Frame Line Format (FLM1) CSV file

Each frame is represented on an individual line capturing only the highest score bounding box of each category detected

All frames are represented exactly once
Designed for videos containing at maximum a single basketball and a single person
Nan values are used with the absence of a detected basketball or person in a frame

clip_ID	width	height	frame	x1_basketball	x2_basketball	y1_basketball	y2_basketball	x1_person	x2_person	y1_person	y2_person

Mock 1 Basketball Tracking Format (BTM1) CSV file

Each line is an individual frame and contains the centerpoint coordinates of the highest scoring basketball detected as well as the radius and "free" column

All frames are represented exactly once
The free column is True if the highest scoring basketballs bounding box has no overlap with the highest scoring persons bounding box
The radius is ((x2 - x1) + (y2 - y1))/2
Nan values are used with the absence of a detected basketball

clip_ID	width	height	frame	x	y	radius	free
int	int	int	int	int	int	float	bool

LabelImg Annotation Format (LI) XML file

link to LabelImg

Each image has a corresponding LI.xml file containing all objects detected

This is used to verify the accuracy of the models detections
Multiple objects are possible for each image

 <annotation>
            <folder></folder>
            <filename></filename>
            <path></path>
            <source>
                <database></database>
            </source>
            <size>
                <width></width>
                <height></height>
                <depth></depth>
            </size>
            <segmented></segmented>
            <object>
                <name></name>
                <pose></pose>
                <truncated></truncated>
                <difficult></difficult>
                <bndbox>
                    <xmin></xmin>
                    <ymin></ymin>
                    <xmax></xmax>
                    <ymax></ymax>
                </bndbox>
            </object>
        </annotation>

Converting between Object Line Format (OL) and LabelImg Annotation Format (LI)

OL	LI
clip_ID	folder
frame	file
width	width
height	height
category	name
score
x1	xmin
x2	xmax
y1	ymin
y2	ymax
model

frame in this repository is the file name minus its extension
score is 100.0 if annotated by a human
model is "human" if annotated by a human

Clip Info Bundel Format (CIB) JSON file

output of image_evaluator

{
	"PATH/TO/FRAME/IMAGE" : 

	{

		"image_path" 		: "PATH/TO/FRAME/IMAGE",
		"image_folder" 		: "IMAGE_FOLDER"
		"image_filename" 	: "IMAGE_FILENAME",
		"image_height" 		: HEIGHT_IN_PIXELS (int),
		"image_width" 		: WIDTH_IN_PIXELS (int),
		"image_items_list" : 

			[
				"category" : "NAME",
				"score" : ACCURACY_SCORE (float),
				"box" : [x1,x2,y1,y2] (ints),
				"model" : "EVALUATION_MODEL"
			]
	}
}

Data Directory Structure

data
│
└───clips
│   |
│   | CLIP_ID1.mp4
|   | CLIP_ID2.mp4
│   | ...
|
└───verified_li_annotations
│   │
│   └───CLIP_ID1
|   |   │
|   |   └───frames
|   |   |   | 1.jpg
|   |   |   | 2.jpg
|   |   |   | ...
|   |   |
|   |   └───li_annotations
|   |   |   | 1.xml
|   |   |   | 2.xml
|   |   |   | ...
|   |   
│   └───CLIP_ID2
│   ...
│   
└───ol_annotations
    │   ol_annotations.csv

ajaykvispute / hi Goto Github PK

hi's Introduction

Shot Detection

AI Basketball Shot Detection and Analysis

Tracking and analysis performed on

Program output world coordinates:

Tracking and analysis performed on

Program output world coordinates:

Tracking and analysis performed on

Program output world coordinates:

Formats for representing video frame object bounding boxes detected

Object Line Format (OL) CSV file

Each object's bounding box in a video recoreded on an individual line

Mock 1 Frame Line Format (FLM1) CSV file

Each frame is represented on an individual line capturing only the highest score bounding box of each category detected

Mock 1 Basketball Tracking Format (BTM1) CSV file

Each line is an individual frame and contains the centerpoint coordinates of the highest scoring basketball detected as well as the radius and "free" column

LabelImg Annotation Format (LI) XML file

Each image has a corresponding LI.xml file containing all objects detected

Converting between Object Line Format (OL) and LabelImg Annotation Format (LI)

Clip Info Bundel Format (CIB) JSON file

output of image_evaluator

Data Directory Structure

hi's People

Contributors

Watchers

Recommend Projects

Recommend Topics

Recommend Org