In the context of this project state-of-the-art real-time deep object detectors are adopted to mobile phones. It is aimed to improve the stand alone performance of these algorithms by using data acquired by mobile sensors.
Current version is designed for TinyYOLOv3 object detector enhanced by the acceleretor data but the algorithm can be easily generalized to other methods.
We have recorded a number of video sequence by mobile phones with different brands. Constructed MSPR-mobile dataset and its groundtruth made open to other parties.
To learn more about the project you can check out the repository.
Dog1 sequence, Target object BB tracked by TinyYOLOv3 (Turquoise), Accelerometer (Red), Integrated (Blue dashed).