Hardware Compression of Video Data

The major Phase 1 objective is to investigate the viability of CSC's innovative spatio-temporal filter energy approach as applied to hardware compression of video data. It is believed that a major advantage of this approach is that STF naturally exploitsthe spatial and temporal frequency and amplitude differences between a target block and its background surround. Coupled with a dynamic programming algorithm, the process extracts and matches the target using a transition cost function that could beadaptively matched to the reference frame. Computational Sensors Corporation (CSC) proposes a program for Hardware Compression of Video Data predicated upon it's core technology of spatio-temporal energy filtering based on the ideally suited computingpower of the Thin Film Analog Image Processor (TAIP) convolution engine. This technology may be enabling in obtaining high quality motion compensated inter-frame coding information in order to overcome the computational complexity and performancelimitations of previous approaches. For example, in MPEG2 this is a very computationally heavy task when searching for the best block e.g. in a 15x15-block move, in some cases, the need arises to perform an exhaustive search over the entire image, say512x512 or larger--this is very computationally intensive.The application of a motion detection approach using Spatial Temporal Filtering based on CSC's unique VLSI chip technology and Dynamic Programming algorithms will enable a new generation ofintelligent systems requiring massive real-time computation with small package size and low power consumption. The development described herein can potentially be extended over a number of diverse detection and identification applications with verysignificant commercial potential. Assuming that TAIP technology could account for an order of magnitude increase in the speed and a corresponding decrease in the power required, the 60% requirement for motion estimation reduces to 6%, with the samecomputation requirements--a 2.5X improvement in battery life. This will become a major enabling issue in MPEG9 combined video and audio-featured products. The TAIP can address the conventional block move/motion estimation problem with significant savings. It also enables a new approach to cue scenes in order to follow moving objects. This is new and conceptually different from that of conventionalblock moves algorithms; with the potential for even greater savings.