基于TMS320C6416的HOG实时优化算法的研究

Real-time Optimization of HOG Based on TMS320C6416: 100fps

HOG算法作为经典的特征描述子被广泛地应用于农业、工业、国防等领域。由于运算量较大,限制了HOG在视频采集终端和便携式设备上的使用和推广。因此,在不影响特征子性能的前提下,基于TI公司1GHz主频的定点处理器TMS320C6416,针对HOG算法的运算效率问题进行了算法层面、代码层面和编译层面的优化,大幅提升HOG算法的效率。以320*256的8位灰度图像为实验对象,进行BIN值为5的HOG特征提取,处理时间为9.1 ms,实现每秒100帧的处理频率,BIN每增加1,处理增加1.5 ms。经过实验验证,优化后的HOG特征仍具有良好的识别性能。

HOG,which is widely applied in the agriculture,industry,national defense field etc.,is a classical descriptor for its excellent performance. However,it is hard to make HOG to meet real-time requirement for its large computation retraining the range of its application such as video and portable equipment. In order to speed up the algorithm while maintaining its performance,the optimized operation is proposed in the field of algorithm,code and compiler based on fixed-point processor TMS320C6416 of TI. The experiment results on the 320* 256 gray images demonstrate that the optimized operation is effective and it achieves 100 fps while the number of BIN equals to five( 9. 1ms),and it would produce an increase of 1. 5ms per BIN. Meanwhile,the performance is still excellent in the embedded system.