Modern video use high resolution and frame rate to achieve high perceptual visual quality. This demand induces significant computation complexity increasing. This work proposed a QP adaptation algorithm for low complexity high efficiency video coding (HEVC) based on a convolutional neural network (CNN) generated header bits map. The proposed algorithm based on a new framework that is consisted of a traditional video encoder and an embedded object detection module with a CNN function. This framework with low complexity and low power consumption has widely demanded in the future. Firstly, the header bits map is generated using a state-of-the-art object detection algorithm, namely you only look once (YOLO). Using the generated header bits map, significant complexity reduction can be achieved by reducing redundant motion estimation for inter prediction of random access mode. Furthermore, an efficient QP adaptation algorithm is proposed based on the map. The simulation results show that the proposed algorithm can achieve 26.8% encoding time saving comparing to the original HEVC algorithm with tiny bit increasing.