The car has increasingly become more and more intelligent throughout the years. Today's radar and vision based safety systems can warn a driver and brake the vehicle automatically if obstacles are detected. Research projects such as the Google Car have even succeeded in creating fully autonomous cars. The demands to obtain the highest rating in safety tests such as Euro NCAP are also steadily increasing, and as a result, the development of these systems have become more attractive for car manufacturers. In the near future, a car must have a system for detecting, and performing automatic braking for pedestrians to receive the highest safety rating of five stars. The prospect is that the volume of active safety system will increase drastically when the car manufacturers start installing them in not only luxury cars, but also in the regularly priced ones. The use of automatic braking comes with a high demand on the performance of active safety systems, false positives must be avoided at all costs. Dollar et al. [2014] introduced Aggregated Channel Features (ACF) which is based on a 10-channel LUV+HOG feature map. The method uses decision trees learned from boosting and has been shown to outperform previous algorithms in object detection tasks. The rediscovery of neural networks, and especially Convolutional Neural Networks (CNN) has increased the performance in almost every field of machine learning, including pedestrian detection. Recently Yang et al.[2015] combined the two approaches by using the the feature maps from a CNN as input to a decision tree based boosting framework. This resulted in state of the art performance on the challenging Caltech pedestrian data set. This thesis presents an approach to improve the performance of a cascade of boosted classifiers by investigating the impact of using color information for pedestrian detection. The color self similarity feature introduced by Walk et al.[2010] was used to create a version better adapted for boosting. This feature is then used in combination with a gradient based feature at the last step of a cascade. The presented feature increases the performance compared to currently used classifiers at Autoliv, on data recorded by Autoliv and on the benchmark Caltech pedestrian data set. / Bilen har genom åren kommit att bli mer och mer intelligent. Dagens radar- och kamerabaserade säkerhetssystem kan varna och bromsa bilen automatiskt om hider detekteras. Forskningsprojekt såsom Google Car har t.o.m lyckats köra bilar helt autonomt. Kraven för att uppnå den högsta säkerhetsklassningen i t.ex. Euro NCAP blir allt strängare i takt med att dessa system utvecklas och som följd har dessa system blivit attraktivare för biltillverkare. Inom en snart framtid kommer det att krävas att en bil har ett system för att upptäcka och att bromsa automatiskt för fotgängare för att uppnå den högsta klassen, fem stjärnor. Förutsikterna är att produktionsvolymer för aktiva säkerhetsytem kommer att öka drastiskt när biltillverkarna börjar utrusta vanliga bilar och inte enbart lyxmodeller med dessa system. Användningen av aktiv bromsning ställer höga krav på prestanda, felakting aktivering av system måste i högsta grad undvikas. Dollar et al. [2014] presenterade Aggregated Channel Features (ACF) som baseras på en tiokanalig LUV+HOG särdragskarta. Metoden använder beslutsträd på pixelnivå som tas fram genom boosting och överträffade tidigare algoritmer för objektigenkänning. Återupptäkten av neurala nätverker och i synnerlighet Convolutional Neural Networks (CNN) har medfört en ökning i prestanda inom nästan alla fält av maskininlärning, inklusive fotgängardetektion. Nyligen kombinerades dessa två metoder av Yang et al.[2015] genom att särdragskartan från ett CNN användes som insignal till ett beslutsträdsbaserat boostingramverk. Detta ledde till det hittills bästa resultatet på det utmanande Caltech pedestrian dataset. I det här examensarbetet presenteras en metod som kan öka prestandan för en kaskad av boostingklassificerare ämnad för fotgängardetektion. Det färgbaserad särdraget color self similarity, Walk et al.[2010], används för att skapa en version som är bättre lämpad för boosting. Det presenterade särdraget ökade prestandan jämfört med befintliga klassificerare som används av Autoliv på både data inspelat av Autoliv och på Caltech pedestrian dataset.

The current technologies for acquisition, storage and transmission of digital data, generate large amounts of data. This quantitative increase is directly proportional to the expansion of multimedia databases, where the bases are part of images. Factors contributing to this expansion is the generation of data access and multimedia, which are frequently used by the population through the media today. Thus, we find a clear need exists for automated systems, capable of dealing with the storage and retrieval of data in a time acceptable to the current standards. To this end, systems are designed for content retrieval of images, where the content is described through its low-level visual features such as shape, texture and color. To have such a system is considered ideal, it must be efficient and effective. The effectiveness will result from the way the information was obtained as a low level of images, considering different conditions of focus, lighting and occlusion. The efficiency is a consequence of the results obtained using the organization of information extracted. The methods of grouping are in one of the useful techniques to reduce the computational complexity of these systems, reducing the computational complexity of the methods implemented, but without losing the representation of information extracted. This work proposes a method for retrieval of images based on content, using appropriate techniques of clustering, a technique for detecting edges and a method to normalize the images in the aspect of enlightenment, to get through it the image descriptors that are robust and can be applied efficiently in a retrieval system for images by content - CBIR (Content Based Image Retrieval). / As tecnologias atuais de aquisição, armazenamento e transmissão de dados digitais geram grandes quantidades de dados. Esse aumento quantitativo é diretamente proporcional à ampliação das bases de dados multimídia, onde se inserem as bases de imagens. Fatores relevantes que contribuem para esta ampliação são o acesso e a geração de dados multimídia, os quais são freqüentemente utilizados pela população através dos meios de comunicação atuais. Desta forma, percebe-se claramente a necessidade existente por sistemas automatizados, capazes de lidar com o armazenamento e a recuperação destes dados em um tempo aceitável para os padrões atuais. Para este fim, são desenvolvidos sistemas de recuperação de imagens por conteúdo, onde este conteúdo é descrito através de suas características visuais de baixo nível, como forma, textura e cor. Para que um sistema deste tipo seja considerado ideal, ele deve ser eficiente e eficaz. A eficácia será resultado da maneira de como foram obtidas as informações de baixo nível das imagens, considerando diferentes condições de foco, oclusão e iluminação. A eficiência é conseqüência dos resultados obtidos utilizando-se a organização das informações extraídas. Os métodos de agrupamento constituem em uma das técnicas úteis para diminuir a complexidade computacional destes sistemas, uma vez que agrupa informações com características semelhantes, sob determinado critério, porém sem perder a representatividade das informações extraídas. Este trabalho propõe um método para recuperação de imagens baseada em conteúdo, que utiliza apropriadamente as técnicas de agrupamento, uma técnica de detecção de cantos e um método para normalizar as imagens no aspecto da iluminação, visando através disso obter descritores da imagem que sejam robustos e possam ser aplicados eficientemente em um sistema de recuperação de imagens por conteúdo - CBIR(Content Based Image Retrieval).

This thesis deals with content based automatic photo categorization. The aim of the work is to experiment with advanced techniques of image represenatation and to create a classifier which is able to process large image dataset with sufficient accuracy and computation speed. A traditional solution based on using visual codebooks is enhanced by computing color features, soft assignment of visual words to extracted feature vectors, usage of image segmentation in process of visual codebook creation and dividing picture into cells. These cells are processed separately. Linear SVM classifier with explicit data embeding is used for its efficiency. Finally, results of experiments with above mentioned techniques of the image categorization are discussed.