Enhancing the Accuracy of ADAS Technology to Improve Driver’s Road Awareness

Semantic Segmentation, Polyline Annotation

The client wanted to detect, localize and categorize all landmarks and pavement defects (pots, cracks, water, etc) on the road to make existing cars safer and facilitate the development of autonomous vehicles. The company needed to develop ADAS technology that recognizes obstacles along the road as well as various road markings. They were looking to increase the accuracy of the model by annotating 60,000 images containing various landmarks and pavement defects such as pots, cracks, water, etc. on the road.  All of these images needed to be annotated with semantic segmentation and polyline annotation, with the highest accuracy possible since the entire project depended on it.

In addition to having a sizable training dataset of 60,000, the client was pressed for time and needed the entire dataset to be annotated within 1.5 months with a quality level of 95%+. A high level of annotation to detail was absolutely paramount since even the smallest scratches needed to be annotated, and all of the lines had to be contoured just right. Any delay in the annotation process would prove to be costly for the client, so they were looking for a data annotation provider whom they could trust to meet deadlines.  

Since the client had already been working with Mindy Support for 3 years and we actualized more than 30 projects together, the client had absolute confidence in Mindy Support’s process, attention to detail, and ability to meet deadlines. Our reputation as one of the industry leaders, as well as our ability to assemble teams quickly, was just what the client was looking for. 

The client was looking to get all of the 60,000 images annotated within a period of 1.5 months, which meant that we only had a few weeks to source and recruit the needed number of candidates to perform the annotation work and get them properly trained. This also included mastering the client’s tool and making sure all annotators were proficient in it. Mindy Support has a detailed process in place on how we go about finding candidates and how the training process is done. This allowed us to source and recruit the candidates of the required profile within strict deadlines.

Our team diligently  annotated all of the lanes and scratches with polylines and other objects with polygons. During the annotation process, we identified many edge cases, thereby helping the client to describe them and expand the project guidelines. It is worth noting that all of this high-quality work was done even though the quality of the images was not very high, so the fact that we were able to achieve the needed quality level shows how much focus and tireless effort our annotators put in. 

Needless to say, the client was very pleased with the results and that our project managers took ownership of the annotation process. This allowed us to complete the project on time without reworking any of the annotations. The client has been working with us for 3 years now with a high level of trust and satisfaction. 

• 60,000 images annotated within 1.5 months
• Achieved a quality level of 95%+
• 3+ years of successful cooperation