Target detection - ADAS practice

ADAS Overview of business scenarios
         Advanced driving AIDS (Advanced Driver Assistance System), abbreviation ADAS, It is installed in the vehicle . Various sensors on the , In the first place - Time to collect environmental data inside and outside the vehicle , To be quiet 、 Identification of dynamic objects 、 Detection and tracking and other technical processing , So that the driver can detect the possible danger in the fastest time , With bow | Active safety technology to pay attention to and improve safety .

Business scenario ：

         Navigation and real-time transportation system TMC、 Electronic police system 、 Car networking 、 Adaptive cruise ACC、 Lane Departure Warning System .、 Lane keeping system 、 Collision avoidance or pre collision system 、 The night vision system 、 Adaptive lighting control 、 Pedestrian protection system 、 Automatic parking system 、 Traffic sign recognition 、 Blind spot detection 、 Driver fatigue detection 、 Downhill control system 、 Electric vehicle alarm system, etc .

Problem description ：

Detect the motor vehicle in the on-board video data 、 Non-Motor Vehicle 、 Pedestrians 、 Traffic identifier
        * Standard object detection problem
                 The appearance of different targets is different
                 light , The effect of occlusion
                 Different perspectives
                 Different sizes
                 Different positions

Data annotation method —— Rectangle label

Judge the performance of the algorithm
        * Detection rate 、 The rate of false positives
                 Only one detection is allowed for each mark
                 Repeated detection will be considered as error detection
        *AP and mAP         

  Dataset resources
        *KITTI Data sets
        *MOT Data sets
        *Berkeley A large-scale automated driving video data set
        *https://blog.csdn.net/u010821 666/article/details/ 79026100

ADAS Scenario industry application status
         Vehicle 、 Non-Motor Vehicle 、 Pedestrian detection is difficult
                * overcast 、 rain 、 Night target detection problem
                * Object detection in crowded scenes
                * Detection problems caused by pedestrian rigid motion
                * Small target detection problem
                * Occlusion problems, etc

ADAS Business scenario data set packaging
KITTI Data set introduction
        *KITTI Data sets
        *http://www.cvlibs.net/datasets/kitti/eval object.php?obj benchmark=2d
        *KITTI The dataset was jointly founded by karlsruhr Institute of technology in Germany and Toyota American Institute of technology , It is the internationally well-known computer vision algorithm evaluation data set under the automatic driving scene .
        * This dataset is used to evaluate stereoscopic images , Optical flow , Visual ranging ,3D Object detection and 3D Tracking and other computer vision technology in the vehicle environment performance .
        *KITTI Including downtown 、 Real image data collected from scenes such as villages and highways , Up to... Per image 15 Cars and 30 A pedestrian , There are also various degrees of occlusion and truncation .
        * The whole dataset consists of 389 Stereo image and optical flow diagram , 39.2 km Visual ranging sequence and more than 200k 3D The image composition of the marked object , With 10Hz Frequency sampling and synchronization .

KITTI Data usage ：

Download link ：http://www.cvlibs.net/datasets/kitti/eval object.php?obj benchmark=2d

 

Label information ：