This deep-learning-based object detection technology is adopted SSD algorithm simplifiedanalyzed by our team.  It is implemented on TI TDA2. With its deep learning library, special EVEDSP on the platform, this system achieves real-time object recognition with 512 x 512 resolution.

With simplified SSD model, we are the first team to implement deep learning on TI TDA2, which can reach 30 fps with 512x512 resolutions. Compared to the one on nVidia Jsetson TX1 / TX2, it can speed up for three timesdecrease 90 hardware cost, which is benefit to be adopted for industry product.

This real-time implementation on embedded platform can be widely applied into edge products, i.e., ADAS, autonomous car, smart surveillance.