整體視覺結(jié)構(gòu)模型及其在道路環(huán)境感知中的應(yīng)用
發(fā)布時(shí)間:2018-07-01 16:48
本文選題:目標(biāo)檢測 + 環(huán)境感知; 參考:《計(jì)算機(jī)工程》2016年10期
【摘要】:智能車輛道路環(huán)境感知中道路目標(biāo)的準(zhǔn)確快速檢測,是智能車輛安全輔助駕駛和自動(dòng)駕駛中的難點(diǎn)。為此,在整體視覺結(jié)構(gòu)模型的基礎(chǔ)上,模擬人體由局部到整體的認(rèn)知機(jī)理,并基于原目標(biāo)局部模型引入結(jié)構(gòu)化信息及概率統(tǒng)計(jì)模型,得到更具姿態(tài)和尺度適應(yīng)性的整體視覺結(jié)構(gòu)模型。該模型在弱標(biāo)注訓(xùn)練樣本的基礎(chǔ)上可實(shí)現(xiàn)目標(biāo)局部特征區(qū)域的自動(dòng)標(biāo)注功能,從而獲得更具特征描述性的目標(biāo)特征。實(shí)驗(yàn)結(jié)果表明,該模型可實(shí)現(xiàn)道路典型目標(biāo)的高檢測率和低誤碼率,且算法效率與經(jīng)典算法相比有所提高。
[Abstract]:Accurate and fast detection of road target in intelligent vehicle road environment perception is a difficult point in intelligent vehicle safety assistant driving and automatic driving. Therefore, based on the global visual structure model, the cognitive mechanism of human body from local to global is simulated, and the structured information and probabilistic statistical model are introduced based on the local model of the original target. A global visual structure model with more adaptability to attitude and scale is obtained. Based on the weakly annotated training samples, the model can automatically label the local feature region of the target, thus obtaining the more descriptive feature of the target. Experimental results show that the model can achieve high detection rate and low bit error rate of typical road targets, and the efficiency of the algorithm is improved compared with the classical algorithm.
【作者單位】: 中國地質(zhì)大學(xué)自動(dòng)化學(xué)院;
【基金】:國家自然科學(xué)基金青年基金資助項(xiàng)目(61503349) 湖北省自然科學(xué)基金資助面上項(xiàng)目(2012FEB6407) 2016年中國地質(zhì)大學(xué)校C類學(xué)術(shù)創(chuàng)新基地開放基金資助項(xiàng)目(AU2015CJ017)
【分類號(hào)】:U463.6;TP391.41
【相似文獻(xiàn)】
相關(guān)期刊論文 前1條
1 張利,汪敏,張建軍,魏中良,尤永;大規(guī)模定制下的汽車底盤產(chǎn)品族結(jié)構(gòu)模型研究[J];合肥工業(yè)大學(xué)學(xué)報(bào)(自然科學(xué)版);2004年12期
,本文編號(hào):2088423
本文鏈接:http://sikaile.net/kejilunwen/qiche/2088423.html
最近更新
教材專著