Success Story Automotive

German Automaker Labels 12M Objects per Annum from Street Videos

Automate the labeling process for street videos

The proliferation of sensors and cameras fosters next-generation ecosystems such as smart cities, smart transportation, smart infrastructure, etc.

The value lies in accurately translating visual data into actionable insights in real-time. This leading German Automaker wanted to automate the labeling process for street videos with live footage from various daytime conditions and locations to identify objects such as vehicles, street lights, and pedestrians.

The challenge

3 key areas

Manually labeling objects from street videos was leading to:

Protracted cycles

Protracted cycles Protracted cycles for object analysis

Increased Costs

Increased Costs Increased cost due to higher manual labor

Poor Engagement

Human Errors Higher scope for human errors

The solution
4 key areas

Apexon brought onboard its Image and Video Analytics platform, a proprietary deep learning platform that speeds up analysis of visual data to enable:

Deep Learning Algorithms

Deep Learning Algorithms

Proactive prediction of objects seen in the video with deep learning algorithms such as EfficientDet, Resnet 50, Mask CNN, and the COCO weights

Object Tracking & computer algorithms

Object Tracking & computer algorithms

Identification of object location across all frames and annotation generation with object tracking algorithms and computer vision

Progression Indicators

Pinhole Camera & Tool Resolution

Conversion of annotations to XMLs with an XML generator to calibrate pinhole camera position and tool resolution

data generator tool

Data generator tool

Curation of tags and corresponding images with XML parser – a data generator tool