DetectionSystem

A system to detect vehicles that are moving based on a series of empirical steps.

The final detections are done using 3 approaches:

• Region proposal + Hand-engineered SVM with more features
• Region proposal + Hand-engineered SVM with lesser features
• Region proposal + YOLOv3

run.sh

A shell script to make execution of detectionsystem.py much simpler and abstracted. The help (-h) parameter can guide you through all the available options.

Sample:

$ sudo chmod +x run.sh
$ ./run.sh -h

Run this script with -r flag to execute all approaches with results validation enabled

Syntax: Template [-c|p|t|f|s|v]

options:

c     0 = Slow SVM, 1 = Fast SVM, 2 = Modified YOLO, 3 = Orginal YOLO
p     0 = For normal processing, 1 = For parallel processing
t     Validate/test results
f     Visualize/Display Frames
s     Visualize/Display Sub-Frames/Regions
v     Verbose mode.
k     Save/keep as file.

Default Option values -c 2 -p 0 -t False -f False -s False -v False

dependencies.sh

Script to check if all the dependencies are downloaded, and download them if not.

Unresolved file dependencies - nuscenes mini/full dataset needs to be downloaded from nuscenes website. Not sharing it or making it availabe through a wgetin dependencies.sh due to Terms of use.

detectionsystem.py

Main file which controls the detections / predictions of all the moving vehicles. Takes in the same arguments as run.sh, and has a similar default option values.

densityscan.py

Clustering module which can cluster features based on different feature distances

SVCmodel.ipynb

File containing the training of hand-engineered SVM classifier models (Fast and Slow) which are done by varying the feature extraction techniques.

requirements.txt

Use the below command to install all the dependencies, or run dependencies.sh for executing it.

$ pip3 install -r requirement.txt