Hailo chooses Visidon to enhance video quality in low-light conditions

Hailo is integrating Visidon’s low-light video enhancement technology into its Hailo-15 AI vision processor to advance AI-powered video analytics for smart cameras, particularly in challenging lighting conditions.

This collaboration is expected to improve video quality, facilitating more accurate detection, recognition, and analysis of subjects in video streams. This advancement is beneficial for applications in biometric surveillance.

With the substantial growth of edge computing in recent years, Hailo anticipates that combining the computational power of the AI processor with advanced video techniques will enable the simultaneous execution of multiple AI tasks.

“We are excited to collaborate with Hailo to enable a remarkable low-light video quality for Hailo-15 AI vision processor-empowered camera devices with our AI de-noise technology,” says Markus Turtinen, CEO of Visidon.

When smart cameras are deployed for surveillance, the capability to execute tasks in parallel will lead to faster detection of objects at high resolutions. This enhancement allows for the accurate identification of smaller and more distant objects, reducing the rate of false alarms.

For example, the CNN-based technology for reducing noise has been demonstrated to improve object detection in real-world scenarios, where noise traditionally impacts the performance of detection models.

One key aspect of improving video quality involves the reduction of noise in image and video inputs. Visidon’s de-noising algorithms play a key role here in enhancing the video footage for better activity monitoring and resulting in clearer, more detailed images.

Visidon has developed a convolution neural network-based technology that makes edge devices ideal for tasks involving image and video analytics. It leverages deep learning to improve clarity and color accuracy in various lighting conditions.

“Not only for improving visual quality but also to increase AI detection accuracy in challenging conditions, offering a real competitive edge for Hailo-15 smart camera customers,” Turtinen adds.

Visidon’s approach to developing advanced noise reduction algorithms is detailed training data, an optimized inference pipeline, camera-specific customization, and fast architecture. This enables the solution to improve noise levels in both objective and subjective manners.

Visidon has collaborated with embedded camera vendors to develop customized algorithms for embedded systems. They are optimized to work with edge device resource constraints without compromising performance.

Particularly with edge devices, power consumption for AI-based video analytics workloads is huge. But Visidon has managed to optimize its network to achieve superior outcomes without necessitating excessive power consumption.

The video enhancement solution’s highlight is its efficiency in low-lighting conditions. The algorithms’ optimization has enabled them to perform better than traditional image signal processor (ISP) technologies, even in ultra-low light below 0.1 lux.

“Our partnership with Visidon is based on a shared belief that the future of ISPs is going to be neural networks based. AI-driven image quality has become standard in smartphones, and we are looking to bring the same level of algorithmic and hardware innovation to smart-cameras,” says Mark Grobman, ML CTO at Hailo.

The co-developed solution is assumed to be capable of processing 4K video streams at up to 60 frames per second in lighting conditions as low as 0.1 lux. Usually, this video format takes a lot of processing power in poor light conditions, but the ability of Visidon models is seen to be an exception.

Also, the Visidon algorithms are designed to be hardware-independent, ensuring that they can be integrated with a wide range of systems. The technology will be showcased at Embedded World in Nürnberg between April 9-11.

“By combining this with Visidon’s proficiency in neural image enhancement, we’ve achieved truly remarkable results,” Grobman continues.

Article Topics

biometrics  |  biometrics at the edge  |  edge AI  |  facial recognition  |  Hailo  |  object recognition  |  video surveillance