hassanali@evrenai.com
+1 (346) 593-7972
hassanali@evrenai.com
+1 (346) 593-7972
Standard Human Activity Recognition (HAR) systems face significant limitations in accuracy and adaptability. To solve this, our internal R&D team embarked on a project to push the boundaries of the field. The result was a novel two-stream deep learning architecture that fuses visual (RGB) and skeletal data, culminating in peer-reviewed research published in the acclaimed journal, Sensors. This work establishes a new state-of-the-art for understanding human actions with unparalleled nuance.
The challenge was to combine RGB video and skeletal data, leveraging both to create a more robust and accurate representation of human activity.
We needed an efficient method to process video data, capturing spatial features and temporal evolution without the high computational cost of traditional 3D-CNNs.
The objective was to develop an LSTM network capable of understanding complex temporal dynamics in skeletal data, using features like joint angles and distances.
“In today's market, true innovation requires a partner who operates at the bleeding edge. Evren AI's published research in Human Activity Recognition is a testament to their deep technical expertise. It is this commitment to fundamental R&D that gives us the confidence to partner with them on our most ambitious and complex AI initiatives.”
Fortune 500 Technology Partner
Engineered a dual-pathway network, using (2+1)D CNN for RGB video and Bidirectional LSTM for skeletal data to capture motion dynamics.
Engineered features based on distances and angles between 17 keypoints, and used Forward Feature Selection to enhance model efficiency and accuracy.
Combined outputs from both streams using fusion techniques, improving prediction accuracy and reliability over individual streams.
The model was tested on the UTD-MHAD dataset, surpassing existing state-of-the-art methods in academic benchmarking.
Our two-stream architecture reached 98.94% accuracy on the UTD-MHAD benchmark, surpassing previous state-of-the-art models.
The novelty of our methodology was validated by the publication of our research in a prestigious, international peer-reviewed journal.
The R&D effort produced a robust framework that Evren AI can now use to address real-world challenges in various industries.
This project highlights our commitment to advancing AI and solving industry challenges, solidifying Evren AI's leadership in applied research.
These results demonstrate the transformative power of AI when implemented strategically and thoughtfully.
Evren AI supports clients across the globe in adopting cutting-edge technology to transform customer interactions.
