首页 > 代码库 > 嵌入式开发之davinci--- 8148/8168/8127 中的添加算饭scd 场景检测
嵌入式开发之davinci--- 8148/8168/8127 中的添加算饭scd 场景检测
Osd
Scd
(1) Introduction
- over view
a) scene change detection
- block diagram
a) graph
b) resvolution
d1:720x576(pal)-25fps 720x480 30-fps(ntsc)--------------704x576 ti
cif:352x288 (支持的处理帧)
quwu:1024x768/4
c) 说明:
The block diagram above illustrates the basic flow of the algorithm. It is helpful to regard the SCD algorithm as an “engine” that consumes input video frames and produces metadata. Input video frames fed to SCD are first partitioned into blocks. Valid YUV input frames provided by the framework are generally CIF resolution or smaller, but block widths are always fixed to be 32-pixels
wide. The xth horizontal and yth vertical block in the partition matrix at time t, i.e. b(x,y,t), is compared against the co-located block from a prior frame b(x,y,t-1) if motion detection for that block is enabled. If frame-level change, e.g. tamper detection, is enabled, then b(x,y,t) is com pared
against a learned model of the scene m(x,y,t-1). The actual operations to generate block-level metadata are more complex than a simple “subtraction,” as depicted in the block diagram.
These metadata are evaluated by logical rules at both a block- and frame-level, depending on whether motion monitoring and/or tamper monitoring is in effect, respectively. Rules for interpreting the metadata are housed inside the algorithm ; however, rules and how they influence decisions can be manipulated by channel-specific parameters selected by the application.
(2) Application programming interface for scene change detection
a) Alglink_scdmod
typedef enum
{
ALG_LINK_SCD_DETECTMODE_DISABLE = 0,
ALG_LINK_SCD_DETECTMODE_MONITOR_FULL_FRAME = 1,
ALG_LINK_SCD_DETECTMODE_MONITOR_BLOCKS = 2,
ALG_LINK_SCD_DETECTMODE_MONITOR_BLOCKS_AND_FRAME = 3
} AlgLink_ScdMode;
ALG_LINK_SCD_DETECTMODE_DISABLE ptz
b) Alglink_scdsensitivity
typedef enum {
ALG_LINK_SCD_SENSITIVITY_VERYLOW = 0,
ALG_LINK_SCD_SENSITIVITY_LOW = 1,
ALG_LINK_SCD_SENSITIVITY_MIDLO = 2,
ALG_LINK_SCD_SENSITIVITY_MID = 3,
ALG_LINK_SCD_SENSITIVITY_MIDHI = 4,
ALG_LINK_SCD_SENSITIVITY_HIGH = 5,
ALG_LINK_SCD_SENSITIVITY_VERYHIGH = 6
} AlgLink_ScdSensitivity;
ALG_LINK_SCD_SENSITIVITY_VERYHIGH
c) Alglink_scdoutput
typedef enum
{
ALG_LINK_SCD_DETECTOR_UNAVAILABLE =-1,
ALG_LINK_SCD_DETECTOR_NO_CHANGE = 0,
ALG_LINK_SCD_DETECTOR_CHANGE = 1
} AlgLink_ScdOutput;
SCD_TI_process
d) Scd struct
e) Scd function call
General guidelines for video and scene characterstics
SCD is designed to analyze video that is acquired from a fixed camera, i.e. the field of view does not change due to panning, tilting, or zooming. Video is processed frame-by-frame. The order and timing of frames are
crucial for analysis. SCD algorithms expect frames to be available in sequential order with inter-frame jitter (variability in frame timing) minimized to be no greater than ±100 ms outside of the specified processing rate.
SCD relies on a fairly stable field of view to identify relevant changes caused by moving objects. Tamper events are assumed to affect the majority of the field of view, depending on the sensitivity setting. To
prevent false alarms and achieve desired results, installers should be careful to position the camera to satisfy the following constraints:
? The video should be in focus and as sharp as possible.
? Camera mounting should be fixed and stable. Excessive vibration or movement from wind, large vehicles, or other external factors should be avoided.
Scene Change Detection API & User’s Guide: Beta 00.50 – January 2012
TI Confidential – NDA Restrictions
? Good contrast with strong edges and corners is desirable for optimum performance. Large reflective surfaces, glare and direct illumination (camera pointed at the sun) can result in poor contrast and must be avoided. Tamper detection in scenes without enough visual texture or contrast, e.g. camera pointed at a blank wall, could be ineffective. However, motion detection has
no similar requirement, except for sufficient illumination.
? No more than 75% of the scene should experience motion or change in appearance at any given time. The size of any individual moving object should not fill more than 50% of the camera’s field of view at any time. If these recommendations are unavoidable and the field of view is easily filled by objects moving into the scene temporarily, e.g. the scene of a camera monitoring traffic is filled by vehicles in traffic, consider lowering the sensitivity to prevent false detections.
? Areas monitored by the camera should be reasonably well lit, e.g. adequate for supporting human eyesight. SCD can work with infrared illuminators to assist in very low-light environments or conditions, but operation under these circumstances can produce undesired effects. Precipitation, e.g. rain drops, snow flakes, ice, sleet, etc., dirt, insects, or other debris on the camera lens can cause SCD algorithms to work improperly.
Adjustments to sensitivity settings will influence detection performance. Lower sensitivity will require a larger degree of high-contrast change for a tamper event to be generated and will result in fewer false events. Higher sensitivity will require a smaller amount of change for an event to be generated and could therefore result in more events, some of them false alarms.
http://blog.csdn.net/mianhuantang848989/article/details/38035731
嵌入式开发之davinci--- 8148/8168/8127 中的添加算饭scd 场景检测