Automating Street Vendors Inspection

The Challenge

Illegal street vendors are known to occupy public spaces, hence interfere public access and mobility. It is a convoluted process in trying to detect those vendors in a predetermined area. Traditionally, officers are carried out to manually monitor the CCTV on a regular basis.

 

Our clients create solutions to overcome problems with the street trader detection using AI system by utilizing public CCTV around the surveillance area. In making this solution, there are several challenges faced in making training data. Starting from many variations of street traders cart that are difficult to categorize and many duplicate images that have exactly the same object and don’t need to be used.



The Solution

Step 1: Image Cleansing

We received many video files sourced from various CCTVs that recorded the street vendor surveillance area. From this video, we split into images and sampling images to reduce the appearance of the same object continuously. With this method we succeeded in reducing unnecessary data annotated by 40%.

CCTV located outdoors tends to have image quality problems such as motion blur, fog and pixel problems. Our platform tried to select images to avoid the above problems and managed to reduce 10% of images that have poor quality.

Step 2: Image Annotation

After receiving the cleaned data, we classify the images based on the time taken using our automated annotation platform.

We classify street vendors object based on the shape of the object to facilitate the process of making bounding box annotations. For example there are those who use carts, tents or cross-legged on the road. For this project, our talented workforce annotated more than 4000 street vendors.

Finally, we make sure the results of the annotation data use multistage quality control so that the data received by the user is in accordance with the quality that we agreed upon earlier.

The Result

Acquaire helps to provide high quality training datas in order to help our clients thrive. Our automated annotation system accelerate  the overall annotation process. 

Detecting illegal street vendors are currently still involve a rigorous manual process. Our Artificial Intelligence technology enables such manual works to be fully automated through the integration of CCTV around the area so that officers could have a more effective reporting system.

Let's discuss about your project

About Us

AcquAIre is a data annotation service from Nodeflux which focuses on producing high quality data for the development of Vision AI technology.

Here for you

(021) 22718184 or hi@nodeflux.io

Where to find us

Jl. Kemang Timur No.24, RT.7/RW.4, Bangka, Mampang Prpt., Jakarta Selatan, 12730