My work

Goal of the Intership

The Goal was to create a context aware video analysis pipeline which can detect and prevent Anomalies in the given environment. Environment could be factory indoors or outdoors, traffic signals and roads, autonomous vehicles, etc.

Unsupervised Frame Prediction

Static Cameras

Siemens Elcita Traffic Camera

Model MSE: 0.000605 Previous Frame MSE: 0.001915

Model is trained on 30 minutes of video.

Siemens Germany Traffic Camera

Model MSE: 0.000104 Previous Frame MSE: 0.000601

Model is trained on 30 minutes of video.

Dashcams

KITTI Self-Driving Car Dashcam

Model MSE: 0.003426 Previous Frame MSE: 0.014377

Model is trained on ~180 minutes of video.

SYNTHIA Virtual Environment Dashcam

Model MSE: 0.008544 Previous Frame MSE: 0.015718

Model is trained on 12 minutes of video.

Path Prediction Algorithm Result

Predictiong the path of the object following given trajectory. Direction of motion is from Dark to light.

Unsupervised Depth Estimation

Object Tracking with Object Detection

Object Detection using YOLO_v3 and overall use is for tracking and Path Prediction