- Mohit Bangale
- August 16, 2023
Ignitarium-Renesas Pretrained AI libraries Release-2
If you found our previous blog ‘Ignitarium Releases Pre-trained AI Applications Library for Renesas RZ/V2L‘ interesting, we are back with the second release of pretrained AI libraries targeted for Renesas RZV2L device. This release consists of new (and more complex!) applications that can readily run on the RZV2L device. Just like the previous set of applications, these applications are open source and free to use for everyone.
Animal Detection Application
This application leverages deep learning to detect an animal from the given input image or video stream from MIPI or USB camera. This application can detect the presence of 12 different animals i.e (Boar, Deer, Crow, Monkey, Bear, Racoon, Fox, Weasel, Skunk, Dog and cat). This application uses YoloV3 object detection network in the background. This application can help in wildlife Conversation, avoid animal accidents (early warning systems), intrusion alert and indoor/outdoor pet monitoring.
Hand Gesture Recognition
Hand gesture recognition application recognizes a hand gesture made by single hand. The deep neural network used in the application identifies 21 key points on the palm. Then, these keypoints are processed using a classification algorithm to recognize the gesture. 9 different gestures such as one, two/scissor, three, four, five/paper, thumbs down, thumbs up and rock are recognized by the app. This application can act as a gesture based remote for devices. There are 2 versions of the same application in this release. One application uses hand keypoints to recognize the gesture, while the second application uses a scene-classification approach to do the same task.
Human Gaze Detection
This application refers to the process of identifying the direction of a person’s gaze or in simple words, where they are looking. Human gaze detection has multiple applications such as driver monitoring systems, Emotion recognition, customer analytics in retail space, digital signage and virtual and augmented reality.
Driver Monitoring System
The driver monitoring system keeps track of the vehicle driver’s behavior and attentiveness on the road during driving. This is a very simple sample application that shows possibilities of a large scale, sophisticated system. The application can tell where the driver is looking at. (Center, Up, Down, Left and Right). The application can also detect eye blinking and yawning activity. This application can have much more complex enhancements like drowsiness detection, distraction detection, talking on phone, beverage drinking, talking to co-passenger etc.
Head count from Top View
Human Head counting application was there in first release. This newer application uses a model that is trained to detect human heads from the top view. This enhancement was done to address the video feeds from camera at public places which are mounted at the ceiling. This helps in occupancy detection at indoor public places, footfall detection, Integration with surveillance and tracking systems, security and monitoring, etc. This application has many challenges as there will be more occlusions with the head at crowded places, varying viewpoints, scale changes, low resolution inputs, etc.
All these sample applications are developed using open-source deep learning models and open-source training data. There are detailed instructions that allow developers to quickly evaluate the existing AI applications on RZ/V2L board as well as to experiment with other types of applications using these Deep Learning models. Due to the inherent limitation of small datasets and state of DL models, the accuracy of the applications may be low. If there is a need to develop a productized version of any of these or other vision AI applications on RZ/V series of AI SoCs, please reach out to the teams at Ignitarium or Renesas, and we will enable the same.
We are excited to see what interesting vision AI applications you would develop on RZ/V2L. Please post your queries using Issues or Discussions sections on GitHub, or write to us at info@ignitarium.com. We would also like to hear about new DL models and applications that you want to see supported on our roadmap.
Stay tuned for more additions to the GitHub page.
Ready Resources:
