Spatiotemporal domain denoising of multi motion detection based on ASIC implementation_data result

The spatiotemporal denoising algorithm for video images is of great significance in filtering out motion noise in video images, improving the encoding efficiency of later video images, and target tracking. Given the real-time requirements of video image acquisition, the spatiotemporal domain denoising algorithm based on ASIC implementation needs to consider the balance between denoising effect and hardware cost. Therefore, this article proposes a spatiotemporal domain denoising algorithm MMD3D based on ASIC implementation for multi motion detection. Firstly, while ensuring the effectiveness of motion detection, the MMD3D algorithm downsamples the current frame image that has undergone NLM spatial denoising by 1\/4 in width and height, and then participates in the calculation of motion detection to increase the equivalent visual window; Secondly, perform point frame difference, block frame difference, and block matching motion detection on the Y\/UV components of the downsampled small image. To improve accuracy, the calculation results of the Y and UV paths are fitted to each other to obtain the final motion weight coefficient (Coef) of the current point; Finally, the Coef of each point in the small image will be upsampled to obtain the Coef of each point in the entire image. Then, the reference frame, current frame, and Coef will be subjected to IIR temporal denoising point by point. At last, the results of IIR temporal denoising and NLM spatial denoising will be fused to output. In the process of ASIC hardware implementation, we adopted a near memory scheme of computing unit and on-chip cache to achieve video image processing 4K@60fps real time processing; The architecture adopts algorithm re-instantiation and Bank SRAM scheme to reduce chip area; By comparison, the MMD3D denoising algorithm in this article can process and significantly improve the noise suppression effect of moving objects in video images in real-time with limited hardware overhead.