Introduction
Advances in computing power is enabling machines to think, understand and act akin to humans. Referred as Artificial Intelligence, this wave of digital transformation is accelerating innovation across industries – and automotive industry no behind. According to a recent industry report, market for automotive AI hardware, software, and services will grow from $404 million in 2016 to $14.0 billion by 2025. What would be the key use-cases driving this growth? Read on to find out:
- What is Artificial intelligence, machine learning, deep learning and more
- How is deep learning – a subset of AI – driving newer applications in automotive industry
- Four applications that are fuelling AI adoption in automotive industry
What is Artificial Intelligence, Machine Learning and Deep Learning?
Any technique that facilitates mimicking human intelligence by logic if-then rules and decision trees is termed as Artificial Intelligence (AI). A subset of AI that includes statistical technique that enables machines to improve at tasks with experience is Machine Learning (ML). This subset has a component in itself which comprises of algorithms that enables software to train itself to perform varied tasks such as voice recognition, face recognition etc. by exposing multi-layered neural network to vast amount of data - this technique is termed Deep Learning (DL).
Figure explaining artificial intelligence and its subsets
Deep Learning comprises of algorithms that work in a manner similar to human brain called Artificial Neural Network (Neural networks are interconnected web of nodes)
- Neural networks- are algorithms that ML uses to model complex patterns. A simple neural network takes an input and passes it through multiple layers of hidden neurons. Output predictions are based on combined input of all neurons.
Deep Learning – A technique for implementing Machine Learning
Deep learning in a broader sense of the term can be understood as developing a process for a system to understand various inputs, classify them and predict results with high accuracy. For instance, something as simple as studying 100 pictures of vegetables, classifying them and eventually system possessing the necessary knowledge in the next stage to be able to identify vegetables from a given image comparing it to the insights gained after an exercise of thoroughly analysing 100 images. The said learning can be supervised, semi supervised or unsupervised.
- Supervised learning - Desired results and cause for them are given as input to the system which now understands what is the expected output for a given input.
- Semi supervised learning - This is the middle ground where in the categories of results are not completely known and systems learn to classify based on available information.
- Unsupervised learning - System is given the output alone and based on that it categorises the input.
4 ways Artificial Intelligence (and Deep Learning) is impacting the Automotive Industry
While there are several applications of Artificial Intelligence or Deep learning in automotive, architecture such as deep neural network has wider applications in segments such as speech recognition, audio recognition, face recognition etc. These applications are applicable not only in autonomous driving but across various aspects of production automation and many more. Four key applications that are finding wide adoption among automotive OEMs include:
- Deep learning in speech recognition and audio recognition
- Deep learning in face recognition
- Deep learning in object detection
- Deep learning in Driver monitoring system (DMS)
Read on to understand them in detail.
1. Deep learning in speech recognition and audio recognition
It is a known fact that when individual calls a large enterprise, it is rare that human answers our call. Instead, an automated device instructs and helps in navigation. This has been present since decades, but DL has made it much more accurate. Now, we can even see cost effective devices which act as our assist systems and have also found importance as a part of home automation. These devices listen to humans and convert it into a machine understandable form. For instance, saying Hello and Helloooooo, both the forms of speech should result in the same recognition. Of course, the sound waves can be quantified, but DL makes it easier for the network to understand and study the pattern better.
2. Deep learning in face recognition
Face recognition is a biometric solution that measures or identifies unique features of a human face. It converts images to pixels for further learning and processing. In face recognition, the network is made to learn from existing and new database of human images. As one can guess, more the number of images for the algorithm to learn, better the accuracy. In automotive applications, Face recognition is used to identify drivers in a fleet of cars, to personalize driver experience and so on. However, developing face recognition system is challenging. To learn more about challenges in face recognition, click here. Also, have look at video demonstration of a low complexity face recognition system for automotive systems here
3. Deep learning in object detection
Autonomous driving is the buzzword in automobile industry this decade and will continue till the next. When it comes to autonomous driving or Advanced Driver Assistance Systems (ADAS), object detection based on vision or cameras plays an important role. Computer Vision (CV) based object detection in DL facilitates a platform for the system to understand various objects that a vehicle may encounter while on the road such as pedestrians, animals, traffic sign, etc. Identifying these signs is the key to an effective autonomous/ADAS technology, hence more developers are using Convolutional Neural Networks (CNN) based DL and other ML techniques to train the system in order to address the problem. Deep learning in object detection can further be looked upon as pedestrian detection and traffic sign recognition. To understand the technology behind object detection, click here.
- Deep learning in pedestrian detection-Pedestrian detection is a long-standing problem not only in automotive use case in ADAS but also in robotics. Pedestrian detection using deep learning needs to be not only real time but also quicker and must consume less computational power.
- Deep learning in traffic sign recognition-Traffic sign recognition has real world application straight out of the box as a part of ADAS. With the availability of off-the-shelf datasets such as GTSRB (German traffic sign) dataset, necessary algorithms can be trained to identify various traffic sign in real world with different climatic condition.
4. Deep learning in Driver monitoring system (DMS)
One of the major factor accounting for accidents across the world is distracted driving. It is estimated that 1 in 5 accidents worldwide is attributed to distracted or drowsy drivers. To overcome this driver monitoring system or driver behaviour monitoring systems have been introduced. This system monitors various facial features and expression to predict the alertness of the driver. But monitoring the features of driver is a herculean task as features varies from one individual to other. To overcome this hurdle, CNN based deep learning models come to rescue.
Conclusion
Since Prof. Geoffery Hinton made a breakthrough in his research on Deep Leaning, the adoption of Artificial Intelligence (and specifically Deep Leaning) has accelerated across industries. Specifically, in Automotive, Deep Learning and Deep Neural Networks are achieving new milestones every single day on varied complex & important scenarios and problems. While challenges surmount in terms of implementing complex Deep Learning models on embedded platforms, getting comprehensive training datasets and delivering accuracy and reliability across diverse working conditions, the benefits of use-cases are too huge to be ignored.
Further reading
Deep learning techniques are all set to transform businesses, in ways never seen before. The use of sophisticated, multi-level deep neural networks is giving businesses inferences, insights and decision-making prowess as advanced as human cognition. PathPartner, with its holistic approach to implementing, optimizing and integrating deep learning methodologies for various applications, is your trusted partner in your deep learning adoption journey.
To know how PathPartner can help in implemented artificial Intelligence and specifically deep learning in your applications, please reach out to us at marcom@pathpartnertech.com
