DESIGN AND IMPLEMENTATION OF A MULTI-CORE PARALLEL LOW ILLUMINATION ENHANCED TARGET DETECTION SYSTEM

In this paper, we propose an embedded deployment method for low illumination enhancement to address vision-based techniques. Low illumination enhancement of pictures, videos and camera images is achieved by multi-core parallel computation using a Phytium multi-core CPU. Target detection is performed on low illumination enhanced images using YOLO-Fastest network and model inference is accelerated by multi-core parallelism. Low illumination enhancement of camera 1920*1080 images and target detection is implemented with 181% multi-core CPU usage and even distribution of usage across cores. After multi-core acceleration for 1920*1080 video, the inference time per video frame is reduced from 570ms to 300ms, which is about 50% speedup.