Fight Detection Project

---

Video_app.online-video-cutter.com.mp4

Video_app.online-video-cutter.com.1.mp4

Abstract

Vision-based activity recognition is one of the most challenging research topics of computer vision and pattern recognition. A specific application of it, namely, detecting fights from surveillance cameras in public areas, prisons, etc., is desired, to get violent incidents under control quickly to reduce casualties.

Results

Model	Top-1 Accuracy	Batch Size (Videos)	Input Frames	Recall (Fight)	Inference Rate (Videos/sec)
r2plus1d_18	77.5%	4	16	0.75	10
r3d_18	82.5%	4	16	0.85	10
mc3_18	90%	4	16	0.80	10
mc3_18	92.5%	8	16	0.90	10
mc3_18	85%	4	32	0.90	5.7

Pytorch Pretrained Models

All pretrained models can be found in this link. https://pytorch.org/vision/stable/models.html

Confusion Matrix

Inference

Run the infer.py script and pass the required arguments (modelPath, streaming, inputPath, outputPath, sequenceLength, skip, showInfo)

python -m infer \
--modelPath="./FDSC/models/model_16_m3_0.8888.pth" \
--streaming=False \
--inputPath="./inputTestVideo.mp4" Or Streaming Url in case of streaming = True \
--outputPath="./outputVideo.mp4" \
--sequenceLength=16 \
--skip=2 \
--showInfo=True

Instructions to Install our Fight Detection Package

Our Package can be found in this link. https://pypi.org/project/Fight-Detection/0.0.3/

1. Install:

pip install Fight-Detection==0.0.3
pip install pytube

2. Download Our Finetuned Model Weights:

import gdown
url = 'https://drive.google.com/uc?id=1MWDeLnpEaZDrKK-OjmzvYLxfjwp-GDcp'
output = 'model_16_m3_0.8888.pth'
gdown.download(url, output, quiet=False)

3. Detect Fight or Not by Pass your Local Video:

from fight_detection import Fight_utils
# Run the Below Function by Input your Test Video Path to get the outPut Video with Fight Detection or Not
Fight_utils.fightDetection(inputPath,seq,skip,outputPath,showInfo)

4. Show the Output Video with Detection:

from moviepy.editor import *
VideoFileClip(outputPath, audio=False, target_resolution=(300,None)).ipython_display()

5- Detect the Fight on Streaming:

Fight_utils.start_streaming(streamingURL)