An Adaptive Video Acquisition Scheme for Object Tracking and Its Performance Optimization

We present a novel adaptive host-chip modular architecture for video acquisition to optimize an overall objective task constrained under given bit rate. The chip is high resolution imaging sensor such as gigapixel focal plane array (FPA) with low computational power deployed on the field remotely, while host server power. communication channel data bandwidth between and accommodate transfer of all captured from chip. performs specific computations also intelligently guides (compress) sent host. This proposed system highly versatile in terms flexibility re-orienting task. In this work, object tracking While our supports any form compression/distortion, paper we use quadtree (QT)-segmented frames. Viterbi (Dynamic Programming) algorithm minimize area normalized weighted rate-distortion allocation resources. receives only these degraded frames analysis. An detector used detect objects, Kalman Filter based tracker track those objects. Evaluation performance done Multiple Object Tracking Accuracy (MOTA) metric. architecture, gains MOTA obtained by twice training different generated distortions 2-step process. Additionally, assisted upscore region proposals further improve performance.