Samsara Inc.

Narendran Rajan

Yan Wang

Justin Joel Delegard

William Evan Welbourne

Naveen Chinya Krishnamurthy

Nathan Hurst

Jing Wang

Greg Allen Cummings

William Nathan John Hurst

Kevin Lai

Ryan Young

Evan Welbourne

Kiren Sekar

Phil Ammirato

Siddharth Heroor

Methods, systems, and apparatuses capture event video segments from an IoT device such as a dashcam in a vehicle. When an even is detected, the original resolution video segment is converted to a low resolution video segment and stored in a cloud backend. Some embodiments thus save bandwidth and storage space by transmitting low resolution video segment to the cloud backend and further save bandwidth and storage space by only storing video in the cloud backend when an event is detected. Bandwidth is also saved because, in some embodiments, original resolution video segments are stored on the IoT device and only transmitted to a reviewer when requested. In the cloud backend, the low resolution video segment is converted to a higher resolution video using a deep learning video upscaler before it is reviewed by a reviewer. A human reviewer can also review the original resolution video segment that is stored on the IoT device if it is desired to see the video segment in its original resolution. In some embodiments, a reviewer can request that an area of interest in the video segment being reviewed is upscaled. In such embodiments the entire low resolution video segment is not always upscaled in its entirety before review.

Optimizing video data upload for dashcam event reviews

A computer-implemented method for machine learning model operation can include, by one or more processors executing program instructions: providing a first training dataset comprising a plurality of images and associated object detection labels, providing a second training dataset comprising a plurality of images and associated classification labels, and providing a machine learning model comprising a model backbone, an object detection task head, and a classification task head. The method can further include training the machine learning model by training the object detection task head using the first training dataset and training the classification task head using the second training dataset. The method can further include deploying the trained machine learning model that includes the trained model backbone and the trained object detection task head but does not include the trained classification task head.

Multi-task machine learning model for event detection

Multiple cameras within a vehicle, such as placed at multiple locations inside and/or outside a bus or semi-truck and trailer, may be chained together to provide a consolidated video stream that may be recorded and/or transmitted wirelessly to a cloud storage device. Rather that each of the cameras independently recording and/or streaming video content, each camera may be configured to deserialize an input video to identify video streams from separate cameras (e.g., such as based on associate virtual channel identifiers) and reserialize its own video stream with the other video streams in the received input. The serialized video stream is then transmitted to a next camera in the daisy chain and/or a video storage device.

Daisy chaining dash cams

Growing community of inventors

San Diego, CA, United States of America

Narendran Rajan